init

.gitignore

@@ -0,0 +1 @@
+*.pyc

COPYING

@@ -0,0 +1,674 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    {one line to give the program's name and a brief idea of what it does.}
+    Copyright (C) {year}  {name of author}
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    {project}  Copyright (C) {year}  {fullname}
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<http://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<http://www.gnu.org/philosophy/why-not-lgpl.html>.

COPYING.LESSER

@@ -0,0 +1,165 @@
+                   GNU LESSER GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+
+  This version of the GNU Lesser General Public License incorporates
+the terms and conditions of version 3 of the GNU General Public
+License, supplemented by the additional permissions listed below.
+
+  0. Additional Definitions.
+
+  As used herein, "this License" refers to version 3 of the GNU Lesser
+General Public License, and the "GNU GPL" refers to version 3 of the GNU
+General Public License.
+
+  "The Library" refers to a covered work governed by this License,
+other than an Application or a Combined Work as defined below.
+
+  An "Application" is any work that makes use of an interface provided
+by the Library, but which is not otherwise based on the Library.
+Defining a subclass of a class defined by the Library is deemed a mode
+of using an interface provided by the Library.
+
+  A "Combined Work" is a work produced by combining or linking an
+Application with the Library.  The particular version of the Library
+with which the Combined Work was made is also called the "Linked
+Version".
+
+  The "Minimal Corresponding Source" for a Combined Work means the
+Corresponding Source for the Combined Work, excluding any source code
+for portions of the Combined Work that, considered in isolation, are
+based on the Application, and not on the Linked Version.
+
+  The "Corresponding Application Code" for a Combined Work means the
+object code and/or source code for the Application, including any data
+and utility programs needed for reproducing the Combined Work from the
+Application, but excluding the System Libraries of the Combined Work.
+
+  1. Exception to Section 3 of the GNU GPL.
+
+  You may convey a covered work under sections 3 and 4 of this License
+without being bound by section 3 of the GNU GPL.
+
+  2. Conveying Modified Versions.
+
+  If you modify a copy of the Library, and, in your modifications, a
+facility refers to a function or data to be supplied by an Application
+that uses the facility (other than as an argument passed when the
+facility is invoked), then you may convey a copy of the modified
+version:
+
+   a) under this License, provided that you make a good faith effort to
+   ensure that, in the event an Application does not supply the
+   function or data, the facility still operates, and performs
+   whatever part of its purpose remains meaningful, or
+
+   b) under the GNU GPL, with none of the additional permissions of
+   this License applicable to that copy.
+
+  3. Object Code Incorporating Material from Library Header Files.
+
+  The object code form of an Application may incorporate material from
+a header file that is part of the Library.  You may convey such object
+code under terms of your choice, provided that, if the incorporated
+material is not limited to numerical parameters, data structure
+layouts and accessors, or small macros, inline functions and templates
+(ten or fewer lines in length), you do both of the following:
+
+   a) Give prominent notice with each copy of the object code that the
+   Library is used in it and that the Library and its use are
+   covered by this License.
+
+   b) Accompany the object code with a copy of the GNU GPL and this license
+   document.
+
+  4. Combined Works.
+
+  You may convey a Combined Work under terms of your choice that,
+taken together, effectively do not restrict modification of the
+portions of the Library contained in the Combined Work and reverse
+engineering for debugging such modifications, if you also do each of
+the following:
+
+   a) Give prominent notice with each copy of the Combined Work that
+   the Library is used in it and that the Library and its use are
+   covered by this License.
+
+   b) Accompany the Combined Work with a copy of the GNU GPL and this license
+   document.
+
+   c) For a Combined Work that displays copyright notices during
+   execution, include the copyright notice for the Library among
+   these notices, as well as a reference directing the user to the
+   copies of the GNU GPL and this license document.
+
+   d) Do one of the following:
+
+       0) Convey the Minimal Corresponding Source under the terms of this
+       License, and the Corresponding Application Code in a form
+       suitable for, and under terms that permit, the user to
+       recombine or relink the Application with a modified version of
+       the Linked Version to produce a modified Combined Work, in the
+       manner specified by section 6 of the GNU GPL for conveying
+       Corresponding Source.
+
+       1) Use a suitable shared library mechanism for linking with the
+       Library.  A suitable mechanism is one that (a) uses at run time
+       a copy of the Library already present on the user's computer
+       system, and (b) will operate properly with a modified version
+       of the Library that is interface-compatible with the Linked
+       Version.
+
+   e) Provide Installation Information, but only if you would otherwise
+   be required to provide such information under section 6 of the
+   GNU GPL, and only to the extent that such information is
+   necessary to install and execute a modified version of the
+   Combined Work produced by recombining or relinking the
+   Application with a modified version of the Linked Version. (If
+   you use option 4d0, the Installation Information must accompany
+   the Minimal Corresponding Source and Corresponding Application
+   Code. If you use option 4d1, you must provide the Installation
+   Information in the manner specified by section 6 of the GNU GPL
+   for conveying Corresponding Source.)
+
+  5. Combined Libraries.
+
+  You may place library facilities that are a work based on the
+Library side by side in a single library together with other library
+facilities that are not Applications and are not covered by this
+License, and convey such a combined library under terms of your
+choice, if you do both of the following:
+
+   a) Accompany the combined library with a copy of the same work based
+   on the Library, uncombined with any other library facilities,
+   conveyed under the terms of this License.
+
+   b) Give prominent notice with the combined library that part of it
+   is a work based on the Library, and explaining where to find the
+   accompanying uncombined form of the same work.
+
+  6. Revised Versions of the GNU Lesser General Public License.
+
+  The Free Software Foundation may publish revised and/or new versions
+of the GNU Lesser General Public License from time to time. Such new
+versions will be similar in spirit to the present version, but may
+differ in detail to address new problems or concerns.
+
+  Each version is given a distinguishing version number. If the
+Library as you received it specifies that a certain numbered version
+of the GNU Lesser General Public License "or any later version"
+applies to it, you have the option of following the terms and
+conditions either of that published version or of any later version
+published by the Free Software Foundation. If the Library as you
+received it does not specify a version number of the GNU Lesser
+General Public License, you may choose any version of the GNU Lesser
+General Public License ever published by the Free Software Foundation.
+
+  If the Library as you received it specifies that a proxy can decide
+whether future versions of the GNU Lesser General Public License shall
+apply, that proxy's public statement of acceptance of any version is
+permanent authorization for you to choose that version for the
+Library.

README.md

@@ -0,0 +1,33 @@
+## Expected Model Output Change (EMOC)
+
+Source code for methods described in the following papers:
+
+- Active learning and discovery of object categories in the presence of unnameable instances
+  C Käding, A Freytag, E Rodner, P Bodesheim, J Denzler
+  Computer Vision and Pattern Recognition (CVPR), 2015
+
+- Large-Scale Active Learning with Approximations of Expected Model Output Changes
+  C Käding, A Freytag, E Rodner, A Perino, J Denzler
+  German Conference on Pattern Recognition (GCPR), 2016
+
+- Watch, Ask, Learn, and Improve: A Lifelong Learning Cycle for Visual Recognition
+  C Käding, E Rodner, A Freytag, J Denzler
+  European Symposium on Artificial Neural Networks (ESANN), 2016
+
+If you use parts of the code, please cite the corresponding papers.
+
+
+##### Dependencies
+
+- Python 2.7
+- numpy
+- scipy
+- scikit-learn
+
+
+##### Usage
+
+1. define setup (see example_setup.cfg)
+2. precompute setup (run evaluation/PrecomputeExperimentalSetup.py setup.cfg)
+3. start experiment (run evaluation/RunExperiment.py setup.cfg)
+4. see results (stored in results.mat)

README.md.bak

@@ -0,0 +1,33 @@
+## Expected Model Output Change (EMOC)
+
+Source code for methods descibed in the following papers:
+
+- Active learning and discovery of object categories in the presence of unnameable instances
+  C Käding, A Freytag, E Rodner, P Bodesheim, J Denzler
+  Computer Vision and Pattern Recognition (CVPR), 2015
+
+- Large-Scale Active Learning with Approximations of Expected Model Output Changes
+  C Käding, A Freytag, E Rodner, A Perino, J Denzler
+  German Conference on Pattern Recognition (GCPR), 2016
+
+- Watch, Ask, Learn, and Improve: A Lifelong Learning Cycle for Visual Recognition
+  C Käding, E Rodner, A Freytag, J Denzler
+  European Symposium on Artificial Neural Networks (ESANN), 2016
+
+If you use parts of the code, please cite the corresponding papers.
+
+
+##### Dependencies
+
+- Python 2.7
+- numpy
+- scipy
+- scikit-learn
+
+
+##### Usage
+
+1. define setup (see example_setup.cfg)
+2. precompute setup (run evaluation/PrecomputeExperimentalSetup.py setup.cfg)
+3. start experiment (run evaluation/RunExperiment.py setup.cfg)
+4. see results (stored in results.mat)

activeLearning/activeLearningGPGenKemoc.py

@@ -0,0 +1,87 @@
+#! /usr/bin/python
+
+import numpy
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+import activeLearningGPGenKprototype
+
+class Classifier(activeLearningGPGenKprototype.ClassifierPrototype):
+
+  def __init__(self,
+               sigmaN = 0.01,
+               useDensity = True,
+               configFile = None):
+
+    activeLearningGPGenKprototype.ClassifierPrototype.__init__(self, sigmaN=sigmaN, configFile=configFile)
+    self.useDensity = helperFunctions.getConfig(configFile, 'activeLearning', 'useDensity', useDensity, 'bool', True)
+
+
+  # x.shape = (number of samples, feat dim)
+  def calcEMOC(self, x, allX=None, kAll=None, sigmaF=None):
+
+    if allX is None:
+      allX = numpy.append(self.X, x, axis=0)
+
+    if kAll is None:
+      kAll = self.kernelFunc(allX,allX.T)
+
+    k = kAll[0:self.X.shape[0],self.X.shape[0]:]
+
+    if sigmaF is None:
+      selfK = numpy.asmatrix(numpy.diag(kAll[self.X.shape[0]:,self.X.shape[0]:])).T
+      sigmaF = self.calcSigmaF(x, k, selfK)
+
+    containsNoise = (self.yUni == -1).any()
+    infY = self.infer(x, containsNoise, k)
+    probs = self.calcProbs(x, infY, sigmaF)
+
+    term1 = 1.0 / (self.sigmaN + sigmaF)
+
+    term2 = numpy.asmatrix(numpy.ones((self.X.shape[0] + 1,x.shape[0]))) * (-1.0)
+    term2[0:self.X.shape[0],:] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float32)*self.sigmaN), k)
+
+    pro = numpy.absolute(infY - 1)
+    contra = numpy.absolute(infY + 1)
+
+    term3 = numpy.repeat(numpy.sum(contra,axis=1),contra.shape[1],axis=1)
+    term3 = numpy.add(numpy.subtract(term3,contra),pro)
+
+    scores = numpy.asmatrix(numpy.zeros((x.shape[0],1)))
+    for cls in range(infY.shape[1]):
+
+      diffAlpha = numpy.multiply(numpy.repeat(numpy.multiply(term1,term3[:,cls]), term2.shape[0], axis=1),term2.T)
+
+      change = numpy.dot(diffAlpha[:,:-1],kAll[0:self.X.shape[0],:])
+      change = numpy.add(change, numpy.multiply(numpy.repeat(diffAlpha[:,-1], kAll.shape[0], axis=1),kAll[self.X.shape[0]:,:]))
+
+      scores = numpy.add(scores, numpy.multiply(probs[:,cls],numpy.sum(numpy.absolute(change),axis=1)))
+
+    return scores
+
+
+  def getDensity(self, sim):
+    return numpy.sum(sim, axis=1) / float(sim.shape[1])
+
+
+  # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    allX = numpy.append(self.X, x, axis=0)
+    kAll = self.kernelFunc(allX, allX)
+
+    k = kAll[0:self.X.shape[0],self.X.shape[0]:]
+    selfK = numpy.asmatrix(numpy.diag(kAll[self.X.shape[0]:,self.X.shape[0]:])).T
+    sigmaF = self.calcSigmaF(x, k, selfK)
+
+    alScores = self.calcEMOC(x, allX=allX, kAll=kAll, sigmaF=sigmaF)
+
+    if self.useDensity:
+
+        density = self.getDensity(kAll[self.X.shape[0]:,:])
+        alScores = numpy.multiply(alScores, density)
+
+    return alScores

activeLearning/activeLearningGPGenKprototype.py

@@ -0,0 +1,282 @@
+import math
+import scipy
+import numpy
+import sklearn.kernel_ridge
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class ClassifierPrototype:
+
+    def __init__(self, sigmaN=0.01, gamma=None, kernel='rbf', configFile=None, probSampleRuns=1000):
+
+        self.sigmaN = helperFunctions.getConfig(configFile, 'activeLearning', 'sigmaN', sigmaN, 'float', True)
+        self.kernel = helperFunctions.getConfig(configFile, 'activeLearning', 'kernel', kernel, 'str', True)
+        self.gamma = helperFunctions.getConfig(configFile, 'activeLearning', 'gamma', gamma, 'float', True)
+        self.numKernelCores = helperFunctions.getConfig(configFile, 'activeLearning', 'numKernelCores', 1, 'int', True)
+        self.probSampleRuns = helperFunctions.getConfig(configFile, 'activeLearning', 'probSampleRuns', probSampleRuns, 'float', True)
+
+        self.K = []
+        self.alpha = []
+        self.X = []
+        self.y = []
+        self.yBin = []
+        self.yUni = []
+
+        self.allowedKernels = ['rbf', 'sigmoid', 'polynomial', 'poly', 'linear', 'cosine']
+
+        if self.kernel not in self.allowedKernels:
+            raise Exception('Unknown kernel %s!'%self.kernel)
+
+
+    def kernelFunc(self, x1, x2=None, gamma=None):
+
+        if gamma is not None:
+            self.gamma = gamma
+
+        if self.kernel in ['rbf']:
+            return numpy.asmatrix(sklearn.kernel_ridge.pairwise_kernels(x1, x2, metric=self.kernel, gamma=self.gamma, n_jobs=self.numKernelCores), dtype=numpy.float)
+        else:
+            return numpy.asmatrix(sklearn.kernel_ridge.pairwise_kernels(x1, x2, metric=self.kernel, n_jobs=self.numKernelCores), dtype=numpy.float)
+
+
+    def checkModel(self):
+
+        if not numpy.all(numpy.isfinite(self.K)):
+            raise Exception('not numpy.all(numpy.isfinite(self.K))')
+
+        if not numpy.all(numpy.isfinite(self.alpha)):
+            raise Exception('not numpy.all(numpy.isfinite(self.alpha))')
+
+        if not numpy.all(numpy.isfinite(self.X)):
+            raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+        if not numpy.all(numpy.isfinite(self.y)):
+            raise Exception('not numpy.all(numpy.isfinite(self.y))')
+
+
+    def train2(self, X, y, sigmaN=None):
+
+        self.train(X[0,:],y[0,:],sigmaN)
+        for idx in range(1,X.shape[0]):
+            self.update(X[idx,:],y[idx,:])
+
+
+    def train(self, X, y, sigmaN=None, gamma=None, kernel=None, numKernelCores=None):
+
+        if sigmaN is not None:
+            self.sigmaN = sigmaN
+
+        if gamma is not None:
+            self.gamma = gamma
+
+        if kernel is not None and kernel in self.allowedKernels:
+            self.kernel = kernel
+
+        if numKernelCores is not None:
+            self.numKernelCores = numKernelCores
+
+        self.X = X
+        self.y = y
+
+        self.yUni = numpy.asmatrix(numpy.unique(numpy.asarray(self.y)))
+
+        tmpVec = numpy.asmatrix(numpy.empty([self.y.shape[0],1]))
+        self.yBin = numpy.asmatrix(numpy.empty([self.y.shape[0],self.yUni.shape[1]]))
+        for cls in range(self.yUni.shape[1]):
+            mask = (self.y == self.yUni[0,cls])
+            tmpVec[mask == True] = 1
+            tmpVec[mask == False] = -1
+            self.yBin[:,cls] = tmpVec
+
+        self.K = numpy.dot(X,X.T)
+        self.alpha = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), self.yBin)
+
+        self.checkModel()
+
+
+    def update(self, x, y):
+
+        # update for known classes
+        binY = numpy.asmatrix(numpy.ones((1,self.alpha.shape[1])))*(-1)
+
+        if self.alpha.shape[1] > 1:
+            binY[0,numpy.asarray(numpy.squeeze(numpy.asarray(y == self.yUni)))] = 1
+        elif y == self.yUni:
+            binY[0,0] = 1
+
+        # get new kernel columns
+        k = self.kernelFunc(self.X, x)
+        selfK = self.getSelfK(x)
+
+        # update alpha
+        term1 = 1.0 / (self.sigmaN + self.calcSigmaF(x, k, selfK).item(0))
+
+        term2 = numpy.asmatrix(numpy.ones((self.X.shape[0] + 1,x.shape[0])), dtype=numpy.float) * (-1.0)
+        term2[0:self.X.shape[0],:] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), k)
+
+        term3 = self.infer(x, k=k) - binY
+
+        self.alpha = numpy.add(numpy.append(self.alpha, numpy.zeros((1,self.alpha.shape[1])), axis=0), numpy.dot(numpy.dot(term1,term2),term3))
+
+        # update K
+        self.K = numpy.append(numpy.append(self.K, k, axis=1), numpy.append(k.T, selfK, axis=1), axis=0)
+
+        # update samples
+        self.X = numpy.append(self.X, x, axis=0)
+        self.y = numpy.append(self.y, y, axis=0)
+
+        # update binary labels for knwon class
+        if (self.yUni == y).any():
+            mask = (y == self.yUni)
+            tmpVec = numpy.empty([self.yBin.shape[1], 1])
+            tmpVec[mask.T == True] = 1
+            tmpVec[mask.T == False] = -1
+            self.yBin = numpy.append(self.yBin, tmpVec.T, axis=0)
+
+        # create labels and alpha for new class
+        else:
+
+            # create bigger matrices
+            tmpyBin = numpy.asmatrix(numpy.empty([self.yBin.shape[0] + 1, self.yBin.shape[1] + 1]))
+            tmpAlpha = numpy.asmatrix(numpy.empty([self.alpha.shape[0], self.alpha.shape[1] + 1]))
+
+            # index of new class
+            tmpIdx = -1
+
+            # check all knwon classes
+            for cls in range(self.yUni.shape[1]):
+
+                # just copy all classes with lower label
+                if y > self.yUni[0,cls]:
+                    tmpyBin[0:-1,cls] = self.yBin[:,cls]
+                    tmpyBin[tmpyBin.shape[0] - 1,cls] = -1
+                    tmpAlpha[:,cls] = self.alpha[:,cls]
+                    tmpIdx = cls
+
+                # copy classes with higher label to shiftet position
+                else: # y < self.yUni[cls]
+                    tmpyBin[0:-1,cls + 1] = self.yBin[:,cls]
+                    tmpyBin[tmpyBin.shape[0] - 1,cls + 1] = -1
+                    tmpAlpha[:,cls + 1] = self.alpha[:,cls]
+
+            # add new binary label vector for new class
+            tmpyBin[0:-1,tmpIdx + 1] = -1
+            tmpyBin[tmpyBin.shape[0] - 1,tmpIdx + 1] = 1
+
+            # set new binary matrix and append new label
+            self.yBin = tmpyBin
+            self.yUni = numpy.sort(numpy.append(self.yUni, y, axis=1))
+
+            # train new alpha for new class
+            tmpAlpha[:,tmpIdx + 1] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), self.yBin[:,tmpIdx + 1])
+            self.alpha = tmpAlpha
+
+        self.checkModel()
+
+
+    # X.shape = (number of samples, feat dim), loNoise = {0,1}
+    def infer(self, x, loNoise=False, k=None):
+
+        if k is None:
+            k = self.kernelFunc(self.X, x)
+
+        loNoise = loNoise and (self.yUni == -1).any()
+
+        pred = numpy.asmatrix(numpy.dot(k.T,self.alpha[:,int(loNoise):]))
+
+        if not numpy.all(numpy.isfinite(pred)):
+            raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+        return pred
+
+
+    # X.shape = (number of samples, feat dim)
+    def test(self, x, loNoise=False):
+
+        loNoise = loNoise and (self.yUni == -1).any()
+
+        return self.yUni[0,numpy.argmax(self.infer(x, loNoise), axis=1) + int(loNoise)]
+
+
+    def calcSigmaF(self, x, k=None, selfK=None):
+
+        if k is None:
+            k = self.kernelFunc(self.X, x)
+
+        if selfK is None:
+            selfK = self.getSelfK(x)
+
+        sigmaF = numpy.subtract(selfK, numpy.sum(numpy.multiply(numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), k),k).T, axis=1))
+
+        if not numpy.all(numpy.isfinite(sigmaF)):
+            raise Exception('not numpy.all(numpy.isfinite(sigmaF))')
+
+        return sigmaF
+
+
+    def getSelfK(self, x):
+
+        selfK = numpy.asmatrix(numpy.empty([x.shape[0],1], dtype=numpy.float))
+        for idx in range(x.shape[0]):
+            selfK[idx,:] = self.kernelFunc(x[idx,:])
+
+        return selfK
+
+
+    # X.shape = (number of samples, feat dim)
+    def calcProbs(self, x, mu=None, sigmaF=None, nmb=None):
+
+        # set number of sampling iterations
+        if nmb is None:
+            nmb = self.probSampleRuns
+
+        # get mu and sigma
+        if mu is None:
+            mu = self.infer(x)
+        if sigmaF is None:
+            sigmaF = self.calcSigmaF(x)
+
+        # prepare
+        probs = numpy.asmatrix(numpy.zeros(mu.shape))
+
+        for idx in range(nmb):
+
+            draws = numpy.asmatrix(numpy.random.randn(mu.shape[0],mu.shape[1]))
+            draws = numpy.add(numpy.multiply(draws, numpy.repeat(sigmaF, draws.shape[1], axis=1)), mu)
+
+            maxIdx = numpy.argmax(draws, axis=1)
+
+            idxs = (range(len(maxIdx)),numpy.squeeze(maxIdx))
+            probs[idxs] = probs[idxs] + 1
+
+        # convert absolute to relative amount
+        return probs/float(nmb)
+
+
+    # x.shape = (feat dim, number of samples)
+    def calcAlScores(self, x):
+
+        return None
+
+
+    # x.shape = (feat dim, number of samples)
+    def getAlScores(self, x):
+
+        alScores = self.calcAlScores(x)
+
+        if not numpy.all(numpy.isfinite(alScores)):
+            raise Exception('not numpy.all(numpy.isfinite(alScores))')
+
+        if alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1:
+            raise Exception('alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1')
+
+        return alScores
+
+
+    # x.shape = (feat dim, number of samples)
+    def chooseSample(self, x):
+
+        return numpy.argmax(self.getAlScores(x), axis=0).item(0)

activeLearning/activeLearningGPLinKemoc.py

@@ -0,0 +1,91 @@
+#! /usr/bin/python
+
+import numpy
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+import activeLearningGPLinKprototype
+
+class Classifier(activeLearningGPLinKprototype.ClassifierPrototype):
+
+  def __init__(self,
+               sigmaN = 0.00178,
+               useDensity = True,
+               configFile = None):
+
+    activeLearningGPLinKprototype.ClassifierPrototype.__init__(self, sigmaN=sigmaN, configFile=configFile)
+    self.useDensity = helperFunctions.getConfig(configFile, 'activeLearning', 'useDensity', useDensity, 'bool', True)
+
+
+  # x.shape = (number of samples, feat dim)
+  def calcEMOC(self, x, allX=None, kAll=None, sigmaF=None):
+
+    if allX is None:
+      allX = numpy.append(self.X, x, axis=0)
+
+    if kAll is None:
+      kAll = numpy.dot(allX,allX.T)
+
+    k = kAll[0:self.X.shape[0],self.X.shape[0]:]
+
+    if sigmaF is None:
+      selfK = numpy.asmatrix(numpy.diag(kAll[self.X.shape[0]:,self.X.shape[0]:])).T
+      sigmaF = self.calcSigmaF(x, k, selfK)
+
+    containsNoise = (self.yUni == -1).any()
+    infY = self.infer(x, containsNoise, k)
+    probs = self.calcProbs(x, infY, sigmaF)
+
+    term1 = 1.0 / (self.sigmaN + sigmaF)
+
+    term2 = numpy.asmatrix(numpy.ones((self.X.shape[0] + 1,x.shape[0]))) * (-1.0)
+    term2[0:self.X.shape[0],:] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float32)*self.sigmaN), k)
+
+    pro = numpy.absolute(infY - 1)
+    contra = numpy.absolute(infY + 1)
+
+    term3 = numpy.repeat(numpy.sum(contra,axis=1),contra.shape[1],axis=1)
+    term3 = numpy.add(numpy.subtract(term3,contra),pro)
+
+    scores = numpy.asmatrix(numpy.zeros((x.shape[0],1)))
+    for cls in range(infY.shape[1]):
+
+      diffAlpha = numpy.multiply(numpy.repeat(numpy.multiply(term1,term3[:,cls]), term2.shape[0], axis=1),term2.T)
+
+      change = numpy.dot(diffAlpha[:,:-1],kAll[0:self.X.shape[0],:])
+      change = numpy.add(change, numpy.multiply(numpy.repeat(diffAlpha[:,-1], kAll.shape[0], axis=1),kAll[self.X.shape[0]:,:]))
+
+      scores = numpy.add(scores, numpy.multiply(probs[:,cls],numpy.sum(numpy.absolute(change),axis=1)))
+
+    return scores
+
+
+  def getDensity(self, sim):
+    return numpy.sum(sim, axis=1) / float(sim.shape[1])
+
+
+  def getDiversity(self, sim):
+    return 1.0 / numpy.max(sim, axis=1)
+
+
+  # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    allX = numpy.append(self.X, x, axis=0)
+    kAll = numpy.dot(allX,allX.T)
+
+    k = kAll[0:self.X.shape[0],self.X.shape[0]:]
+    selfK = numpy.asmatrix(numpy.diag(kAll[self.X.shape[0]:,self.X.shape[0]:])).T
+    sigmaF = self.calcSigmaF(x, k, selfK)
+
+    alScores = self.calcEMOC(x, allX=allX, kAll=kAll, sigmaF=sigmaF)
+
+    if self.useDensity:
+
+        density = self.getDensity(kAll[self.X.shape[0]:,:])
+        alScores = numpy.multiply(alScores, density)
+
+    return alScores

activeLearning/activeLearningGPLinKprototype.py

@@ -0,0 +1,250 @@
+#! /usr/bin/python
+
+import numpy
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class ClassifierPrototype:
+
+  def __init__(self, sigmaN = 0.00178, configFile=None, probSampleRuns=1000):
+
+    self.sigmaN = helperFunctions.getConfig(configFile, 'activeLearning', 'sigmaN', sigmaN, 'float', True)
+    self.probSampleRuns = helperFunctions.getConfig(configFile, 'activeLearning', 'probSampleRuns', probSampleRuns, 'float', True)
+
+    self.K = []
+    self.alpha = []
+    self.X = []
+    self.y = []
+    self.yBin = []
+    self.yUni = []
+
+
+  def checkModel(self):
+
+    if not numpy.all(numpy.isfinite(self.K)):
+      raise Exception('not numpy.all(numpy.isfinite(self.K))')
+
+    if not numpy.all(numpy.isfinite(self.alpha)):
+      raise Exception('not numpy.all(numpy.isfinite(self.alpha))')
+
+    if not numpy.all(numpy.isfinite(self.X)):
+      raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+    if not numpy.all(numpy.isfinite(self.y)):
+      raise Exception('not numpy.all(numpy.isfinite(self.y))')
+
+
+  def train2(self, X, y, sigmaN=None):
+
+    self.train(X[0,:],y[0,:],sigmaN)
+    for idx in range(1,X.shape[0]):
+      self.update(X[idx,:],y[idx,:])
+
+
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y, sigmaN=None):
+
+    if sigmaN is not None:
+        self.sigmaN = sigmaN
+
+    self.X = X
+    self.y = y
+
+    self.yUni = numpy.asmatrix(numpy.unique(numpy.asarray(self.y)))
+
+    tmpVec = numpy.asmatrix(numpy.empty([self.y.shape[0],1]))
+    self.yBin = numpy.asmatrix(numpy.empty([self.y.shape[0],self.yUni.shape[1]]))
+    for cls in range(self.yUni.shape[1]):
+      mask = (self.y == self.yUni[0,cls])
+      tmpVec[mask == True] = 1
+      tmpVec[mask == False] = -1
+      self.yBin[:,cls] = tmpVec
+
+    self.K = numpy.dot(X,X.T)
+    self.alpha = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), self.yBin)
+
+    self.checkModel()
+
+
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+
+    # update for known classes
+    binY = numpy.asmatrix(numpy.ones((1,self.alpha.shape[1])))*(-1)
+
+    if self.alpha.shape[1] > 1:
+      binY[0,numpy.asarray(numpy.squeeze(numpy.asarray(y==self.yUni)))] = 1
+    elif y==self.yUni:
+        binY[0,0] = 1
+
+    # get new kernel columns
+    k = numpy.dot(self.X,x.T)
+    selfK = numpy.sum(numpy.multiply(x,x), axis=1)
+
+    # get score of new sample
+    infY = self.infer(x, k=k)
+
+    # update alpha
+    term1 = 1.0 / (self.sigmaN + self.calcSigmaF(x, k, selfK).item(0))
+
+    term2 = numpy.asmatrix(numpy.ones((self.X.shape[0] + 1,x.shape[0])), dtype=numpy.float)*(-1.0)
+    term2[0:self.X.shape[0],:] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), k)
+
+    term3 = infY - binY
+
+    self.alpha = numpy.add(numpy.append(self.alpha, numpy.zeros((1,self.alpha.shape[1])), axis=0), numpy.dot(numpy.dot(term1,term2),term3))
+
+    # update K
+    self.K = numpy.append(numpy.append(self.K, k, axis=1), numpy.append(k.T, selfK, axis=1), axis=0)
+
+    # update samples
+    self.X = numpy.append(self.X, x, axis=0)
+    self.y = numpy.append(self.y, y, axis=0)
+
+    # update binary labels for knwon class
+    if (self.yUni == y).any():
+      mask = (y == self.yUni)
+      tmpVec = numpy.empty([self.yBin.shape[1], 1])
+      tmpVec[mask.T == True] = 1
+      tmpVec[mask.T == False] = -1
+      self.yBin = numpy.append(self.yBin, tmpVec.T, axis=0)
+
+    # create labels and alpha for new class
+    else:
+
+      # create bigger matrices
+      tmpyBin = numpy.asmatrix(numpy.empty([self.yBin.shape[0] + 1, self.yBin.shape[1] + 1]))
+      tmpAlpha = numpy.asmatrix(numpy.empty([self.alpha.shape[0], self.alpha.shape[1] + 1]))
+
+      # index of new class
+      tmpIdx = -1
+
+      # check all knwon classes
+      for cls in range(self.yUni.shape[1]):
+
+        # just copy all classes with lower label
+        if y > self.yUni[0,cls]:
+          tmpyBin[0:-1,cls] = self.yBin[:,cls]
+          tmpyBin[tmpyBin.shape[0] - 1,cls] = -1
+          tmpAlpha[:,cls] = self.alpha[:,cls]
+          tmpIdx = cls
+
+        # copy classes with higher label to shiftet position
+        else: # y < self.yUni[cls]
+          tmpyBin[0:-1,cls + 1] = self.yBin[:,cls]
+          tmpyBin[tmpyBin.shape[0] - 1,cls + 1] = -1
+          tmpAlpha[:,cls + 1] = self.alpha[:,cls]
+
+      # add new binary label vector for new class
+      tmpyBin[0:-1,tmpIdx + 1] = -1
+      tmpyBin[tmpyBin.shape[0] - 1,tmpIdx + 1] = 1
+
+      # set new binary matrix and append new label
+      self.yBin = tmpyBin
+      self.yUni = numpy.sort(numpy.append(self.yUni, y, axis=1))
+
+      # train new alpha for new class
+      tmpAlpha[:,tmpIdx + 1] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), self.yBin[:,tmpIdx + 1])
+      self.alpha = tmpAlpha
+
+    self.checkModel()
+
+
+  # X.shape = (number of samples, feat dim), loNoise = {0,1}
+  def infer(self, x, loNoise=False, k=None):
+
+    if k is None:
+      k = numpy.dot(self.X,x.T)
+
+    loNoise = loNoise and (self.yUni == -1).any()
+
+    pred = numpy.asmatrix(numpy.dot(k.T,self.alpha[:,int(loNoise):]))
+
+    if not numpy.all(numpy.isfinite(pred)):
+      raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+    return pred
+
+
+  # X.shape = (number of samples, feat dim)
+  def test(self, x, loNoise=False):
+
+    loNoise = loNoise and (self.yUni == -1).any()
+
+    return self.yUni[0,numpy.argmax(self.infer(x, loNoise), axis=1) + int(loNoise)]
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcSigmaF(self, x, k=None, selfK=None):
+
+    if k is None:
+        k = numpy.dot(self.X,x.T)
+
+    if selfK is None:
+        selfK = numpy.sum(numpy.multiply(x,x), axis=1)
+
+    sigmaF = numpy.subtract(selfK, numpy.sum(numpy.multiply(numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), k),k).T, axis=1))
+
+    if not numpy.all(numpy.isfinite(sigmaF)):
+      raise Exception('not numpy.all(numpy.isfinite(sigmaF))')
+
+    return sigmaF
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcProbs(self, x, mu=None, sigmaF=None, nmb=None):
+
+    # set number of sampling iterations
+    if nmb is None:
+        nmb = self.probSampleRuns
+
+    # get mu and sigma
+    if mu is None:
+      mu = self.infer(x)
+    if sigmaF is None:
+      sigmaF = self.calcSigmaF(x)
+
+    # prepare
+    probs = numpy.asmatrix(numpy.zeros(mu.shape))
+
+    for idx in range(nmb):
+
+      draws = numpy.asmatrix(numpy.random.randn(mu.shape[0],mu.shape[1]))
+      draws = numpy.add(numpy.multiply(draws, numpy.repeat(sigmaF, draws.shape[1], axis=1)), mu)
+
+      maxIdx = numpy.argmax(draws, axis=1)
+
+      idxs = (range(len(maxIdx)),numpy.squeeze(maxIdx))
+      probs[idxs] = probs[idxs] + 1
+
+    # convert absolute to relative amount
+    return probs/float(nmb)
+
+
+  # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    return None
+
+
+  # x.shape = (feat dim, number of samples)
+  def getAlScores(self, x):
+
+    alScores = self.calcAlScores(x)
+
+    if not numpy.all(numpy.isfinite(alScores)):
+      raise Exception('not numpy.all(numpy.isfinite(alScores))')
+
+    if alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1:
+      raise Exception('alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1')
+
+    return alScores
+
+
+  # x.shape = (feat dim, number of samples)
+  def chooseSample(self, x):
+
+    return numpy.argmax(self.getAlScores(x), axis=0).item(0)

activeLearning/activeLearningLinGPemoc.py

@@ -0,0 +1,75 @@
+#! /usr/bin/python
+
+import numpy
+import pickle
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+import activeLearningLinGPprototype
+
+class Classifier(activeLearningLinGPprototype.ClassifierPrototype):
+
+  def __init__(self,
+                    sigmaN = 0.00178,
+                    usePde = True,
+                    configFile=None):
+
+    activeLearningLinGPprototype.ClassifierPrototype.__init__(self, sigmaN=sigmaN, configFile=configFile)
+    self.usePde = helperFunctions.getConfig(configFile, 'activeLearning', 'usePde', usePde, 'bool', True)
+
+
+  # x.shape = (number of samples, feat dim)
+  def calcEMOC(self, x, allX=None):
+
+    if allX is None:
+      allX = numpy.append(self.X, x, axis=0)
+
+    containsNoise = (self.yUni == -1).any()
+
+    tmpVec1 = self.invCreg*x.T
+    tmpVec2 = numpy.sum(numpy.multiply(x.T,tmpVec1), axis=0)
+
+    sigmaF = self.calcSigmaF(x, tmpVec2)
+
+    infY = self.infer(x, containsNoise)
+    probs = self.calcProbs(x, infY, sigmaF)
+
+    term1 = 1.0/(1.0 + tmpVec2)
+
+    term2 = numpy.sum(numpy.absolute(allX*tmpVec1), axis=0)
+
+    pro = numpy.absolute(infY - 1)
+    contra = numpy.absolute(infY + 1)
+    diff = numpy.repeat(numpy.sum(contra,axis=1),contra.shape[1],axis=1)
+    diff = (diff - contra + pro)
+
+    term3 = numpy.sum(numpy.multiply(probs,diff),axis=1)
+
+    res = numpy.multiply(numpy.multiply(term1, term2).T, term3)
+
+    return numpy.multiply(numpy.multiply(term1, term2).T, term3)
+
+
+  # x.shape = (number of samples, feat dim)
+  def calcEMOCpde(self, x):
+
+    allX = numpy.append(self.X, x, axis=0)
+
+    scores = self.calcEMOC(x, allX)
+
+    density = x*numpy.mean(allX, axis=0).T
+
+    return numpy.multiply(scores, density)
+
+
+   # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    if self.usePde:
+      return self.calcEMOCpde(x)
+    else:
+      return self.calcEMOC(x)
+

activeLearning/activeLearningLinGPemocApprox.py

@@ -0,0 +1,163 @@
+#! /usr/bin/python
+
+import numpy
+import scipy.cluster.vq
+import pickle
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+import activeLearningLinGPprototype
+
+class Classifier(activeLearningLinGPprototype.ClassifierPrototype):
+
+  def __init__(self,
+                    sigmaN = 0.00178,
+                    usePde = True,
+                    configFile=None):
+
+    activeLearningLinGPprototype.ClassifierPrototype.__init__(self, sigmaN=sigmaN, configFile=configFile)
+    self.usePde = helperFunctions.getConfig(configFile, 'activeLearning', 'usePde', usePde, 'bool', True)
+    self.approxMode = helperFunctions.getConfig(self.configFile, 'activeLearning', 'approxMode', 'rnd', 'str', True)
+    self.approxSize = helperFunctions.getConfig(self.configFile, 'activeLearning', 'approxSize', 500, 'int', True)
+
+    self.cachedClusters = None
+    self.cachedApprox = None
+    self.cachedDensity = None
+    
+
+  # x.shape = (number of samples, feat dim)
+  def calcEMOC(self, x, allX=None, density=None):
+
+    containsNoise = (self.yUni == -1).any()
+
+    tmpVec1 = self.invCreg*x.T
+    tmpVec2 = numpy.sum(numpy.multiply(x.T,tmpVec1), axis=0)
+
+    sigmaF = self.calcSigmaF(x, tmpVec2)
+
+    infY = self.infer(x, containsNoise)
+    probs = self.calcProbs(x, infY, sigmaF)
+
+    term1 = 1.0/(1.0 + tmpVec2)
+
+    approxSum = self.approxSum(x, allX, tmpVec1, density)
+
+    pro = numpy.absolute(infY - 1)
+    contra = numpy.absolute(infY + 1)
+    diff = numpy.repeat(numpy.sum(contra,axis=1),contra.shape[1],axis=1)
+    diff = (diff - contra + pro)
+
+    term3 = numpy.sum(numpy.multiply(probs,diff),axis=1)
+
+    res = numpy.multiply(numpy.multiply(term1, approxSum).T, term3)
+
+    return numpy.multiply(numpy.multiply(term1, self.approxSum(x, allX, tmpVec1, density)).T, term3)
+
+
+  # x.shape = (number of samples, feat dim)
+  def calcEMOCpde(self, x, allX=None):
+
+    if allX is None:
+        allX = numpy.append(self.X, x, axis=0)
+
+    if self.cachedDensity is None:
+        density = self.calcDensity(x, allX)
+    else:
+        density = self.cachedDensity
+
+    scores = self.calcEMOC(x, allX, density)
+
+    return numpy.multiply(scores, density)
+
+
+   # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    allX = numpy.append(self.X, x, axis=0)
+
+    if self.usePde:
+        return self.calcEMOCpde(x, allX)
+    else:
+        return self.calcEMOC(x, allX)
+
+
+  def approxSum(self, x, allX=None, invCregDotX=None, density=None):
+
+    if self.approxMode == 'Call':
+        return self.cachedApprox
+
+    ###
+
+    if invCregDotX is None:
+        invCregDotX = self.invCreg*x.T
+
+    ###
+
+    if self.approxMode == 'clustering':
+        return numpy.sum(numpy.absolute(self.cachedClusters*invCregDotX), axis=0)
+
+    ###
+
+    amount = min(self.approxSize, self.X.shape[0] + x.shape[0])
+
+    if allX is None:
+        allX = numpy.append(self.X, x, axis=0)
+
+    ###
+
+    if self.approxMode == 'rnd':
+        return numpy.sum(numpy.absolute(allX[numpy.random.permutation(allX.shape[0])[:amount], :]*invCregDotX), axis=0)
+
+    else:
+        raise Exception('Approximation mode > %s < unknown!'%self.approxMode)
+
+
+  def prepareApprox(self, x):
+
+    allX = numpy.append(self.X, x, axis=0)
+
+    ###
+
+    if self.usePde:
+        self.cachedDensity = self.calcDensity(x, allX)
+
+    ###
+
+    if self.approxMode == 'rnd' or self.approxMode == 'knn':
+        return
+
+    ###
+
+    if self.approxMode == 'clustering':
+        amount = min(self.approxSize, allX.shape[0])
+        self.cachedClusters = scipy.cluster.vq.kmeans(allX, amount)[0]
+        print 'data points:', allX.shape[0], ', cluster requested:', amount, ', cluster found:', self.cachedClusters.shape[0]
+        return
+
+    else:
+        raise Exception('Approximation mode > %s < unknown!'%self.approxMode)
+
+
+  def clearApprox(self, idx=None):
+
+    if idx is None:
+        self.cachedApprox = None
+        self.cachedClusters = None
+        if self.usePde:
+            self.cachedDensity = None
+    else:
+        if self.cachedApprox is not None:
+            self.cachedApprox = numpy.delete(self.cachedApprox, (idx), axis=0)
+        if self.usePde and self.cachedDensity is not None:
+            self.cachedDensity = numpy.delete(self.cachedDensity, (idx), axis=0)
+
+
+  def calcDensity(self, x, allX=None):
+
+    if allX is None:
+        allX = numpy.append(self.X, x, axis=0)
+
+    return x*numpy.mean(allX, axis=0).T

activeLearning/activeLearningLinGPprototype.py

@@ -0,0 +1,185 @@
+#! /usr/bin/python
+
+import numpy
+import pickle
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class ClassifierPrototype:
+
+  def __init__(self, sigmaN = 0.00178, configFile=None):
+
+    self.sigmaN = helperFunctions.getConfig(configFile, 'activeLearning', 'sigmaN', sigmaN, 'float', True)
+
+    self.invCreg = []
+    self.X = []
+    self.yBin = []  # .shape = (number of samples, number of unique classes)
+    self.yUni = []
+    self.w = []  # .shape = (feat dim, number of unique classes)
+
+
+  def checkModel(self):
+
+    if not numpy.all(numpy.isfinite(self.invCreg)):
+      raise Exception('not numpy.all(numpy.isfinite(self.invCreg))')
+
+    if not numpy.all(numpy.isfinite(self.w)):
+      raise Exception('not numpy.all(numpy.isfinite(self.w))')
+
+    if not numpy.all(numpy.isfinite(self.X)):
+      raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+    if not numpy.all(numpy.isfinite(self.yUni)):
+      raise Exception('not numpy.all(numpy.isfinite(self.yUni))')
+
+    if len(numpy.unique(self.y)) > 2:
+      raise Exception('len(numpy.unique(self.y)) > 2')
+
+
+  def train2(self, X, y, sigmaN=None):
+
+    self.train(X[0,:],y[0,:],sigmaN)
+    for idx in range(1,X.shape[0]):
+      self.update(X[idx,:],y[idx,:])
+
+
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y, sigmaN=None):
+
+    if sigmaN is not None:
+        self.sigmaN = sigmaN
+
+    # get all known classes
+    self.yUni = numpy.asmatrix(numpy.unique(numpy.asarray(y)))
+
+    # save stuff
+    self.X = X
+
+    # calculate inverse of C_reg
+    self.invCreg = numpy.linalg.inv((X.T*X) + numpy.identity(X.shape[1])*self.sigmaN)
+
+    # get binary labels for each ovr - classifier
+    self.yBin = numpy.asmatrix(numpy.empty((y.shape[0],self.yUni.shape[1])), dtype=numpy.int)
+    for cls in range(self.yUni.shape[1]):
+      self.yBin[:,cls] = 2*(y == self.yUni[0,cls]) - 1
+
+    # calculate w
+    self.w = self.invCreg*X.T*self.yBin
+
+    self.checkModel()
+
+
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+
+    # update w and C_reg
+    tmpVec = self.invCreg*x.T
+    tmpScalar = 1.0 + x*tmpVec
+    self.w = self.w + tmpVec*((numpy.asmatrix(2*(y==self.yUni) - 1) - x*self.w)/tmpScalar)
+    self.invCreg = self.invCreg - ((tmpVec*tmpVec.T)/tmpScalar)
+
+    # update samples
+    self.X = numpy.append(self.X, x, axis=0)
+
+    # update binary labels for known class
+    self.yBin = numpy.append(self.yBin, 2*(y == self.yUni) - 1, axis=0)
+
+    # create labels and w for new class
+    if not (self.yUni == y).any():
+
+        # get insertion idx
+        idx = numpy.searchsorted(numpy.ravel(numpy.asarray(self.yUni)), y[0,0])
+
+        # store new label and find new class index
+        self.yUni = numpy.insert(self.yUni, [idx], y, axis=1)
+
+        ## add new binary label vector for new class
+        self.yBin = numpy.insert(self.yBin, [idx], numpy.zeros((self.yBin.shape[0],1), dtype=numpy.int), axis=1)
+        self.yBin[:-1,idx] = -1
+        self.yBin[-1,idx] = 1
+
+        ## train new w for new class
+        self.w = numpy.insert(self.w, [idx], self.invCreg*self.X.T*self.yBin[:,idx], axis=1)
+
+    self.checkModel()
+
+
+  # X.shape = (number of samples, feat dim), loNoise = {0,1}
+  def infer(self, x, loNoise=False):
+
+    loNoise = loNoise and (self.yUni == -1).any()
+
+    # division by number of training samples is a hack to prevent huge scores
+    return numpy.asmatrix(x*self.w[:,int(loNoise):]) #/ float(self.X.shape[0])
+
+
+  # X.shape = (number of samples, feat dim)
+  def test(self, x, loNoise=False):
+
+    loNoise = loNoise and (self.yUni == -1).any()
+
+    return self.yUni[0,numpy.argmax(self.infer(x, loNoise), axis=1) + int(loNoise)]
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcSigmaF(self, x, tmpVec=None):
+
+    if tmpVec is None:
+        return numpy.sum(numpy.multiply(x,(self.sigmaN*self.invCreg*x.T).T),axis=1)
+    else:
+        return self.sigmaN*tmpVec.T
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcProbs(self, x, mu=None, sigmaF=None, nmb=1000):
+
+    # get mu and sigma
+    if mu is None:
+      mu = self.infer(x)
+    if sigmaF is None:
+      sigmaF = self.calcSigmaF(x)
+
+    # prepare
+    probs = numpy.asmatrix(numpy.zeros(mu.shape))
+
+    for idx in range(nmb):
+
+      draws = numpy.asmatrix(numpy.random.randn(mu.shape[0],mu.shape[1]))
+      draws = numpy.multiply(draws, numpy.repeat(sigmaF, draws.shape[1], axis=1)) + mu
+
+      maxIdx = numpy.argmax(draws, axis=1)
+
+      idxs = (range(len(maxIdx)),numpy.squeeze(maxIdx))
+      probs[idxs] = probs[idxs] + 1
+
+    # convert absolute to relative amount
+    return probs/float(nmb)
+
+
+  # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    return None
+
+
+  # x.shape = (feat dim, number of samples)
+  def getAlScores(self, x):
+
+    alScores = self.calcAlScores(x)
+
+    if not numpy.all(numpy.isfinite(alScores)):
+      raise Exception('not numpy.all(numpy.isfinite(alScores))')
+
+    if alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1:
+      raise Exception('alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1')
+
+    return alScores
+
+
+  # x.shape = (feat dim, number of samples)
+  def chooseSample(self, x):
+
+    return numpy.argmax(self.getAlScores(x), axis=0).item(0)

activeLearning/activeLearningWlinGP1vs2.py

@@ -0,0 +1,67 @@
+#! /usr/bin/python
+
+import numpy
+import pickle
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+import activeLearningWlinGPprototype
+
+class Classifier(activeLearningWlinGPprototype.ClassifierPrototype):
+
+    def __init__(self,
+                 sigmaN = 0.00178,
+                 useDiversity = False,
+                 useDensity = False,
+                 useVariance = False,
+                 loNoise = True,
+                 configFile=None):
+
+        activeLearningWlinGPprototype.ClassifierPrototype.__init__(self, sigmaN=sigmaN, configFile=configFile)
+        self.useDiversity = helperFunctions.getConfig(configFile, 'activeLearning', 'useDiversity', useDiversity, 'bool', True)
+        self.useVariance = helperFunctions.getConfig(configFile, 'activeLearning', 'useVariance', useVariance, 'bool', True)
+        self.useDensity = helperFunctions.getConfig(configFile, 'activeLearning', 'useDensity', useDensity, 'bool', True)
+        self.loNoise = helperFunctions.getConfig(configFile, 'activeLearning', 'loNoise', loNoise, 'bool', True)
+
+
+    def getDensity(self, sim):
+        return numpy.sum(sim, axis=1) / float(sim.shape[1])
+
+
+    def getDiversity(self, sim):
+        return 1.0 / numpy.max(sim, axis=1)
+
+
+    # x.shape = (feat dim, number of samples)
+    def calcAlScores(self, x):
+
+        loNoise = (self.yUni == -1).any() and self.loNoise
+        sortedScores = numpy.sort(self.infer(x, loNoise), axis=1)
+        alScores = numpy.absolute(sortedScores[:,-1] - sortedScores[:,-2])*(-1.0)
+        sim = None
+
+        if self.useDensity:
+
+            sim = numpy.dot(x, numpy.append(self.X, x, axis=0).T)
+            density = self.getDensity(sim)
+            alScores = numpy.multiply(alScores, density)
+
+        elif self.useDiversity:
+
+            if sim is None:
+                sim = numpy.dot(x, self.X.T)
+                diversity = self.getDiversity(sim)
+            else:
+                diversity = self.getDiversity(sim[:,:self.X.shape[0]])
+
+            alScores = numpy.multiply(alScores, diversity)
+
+        elif self.useVariance:
+
+            variance = self.calcSigmaF(x)
+            alScores = numpy.multiply(alScores, variance)
+
+        return alScores

activeLearning/activeLearningWlinGPprototype.py

@@ -0,0 +1,204 @@
+#! /usr/bin/python
+
+import numpy
+import pickle
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class ClassifierPrototype:
+
+  def __init__(self,
+                    sigmaN = 0.00178,
+                    configFile=None):
+
+    self.sigmaN = helperFunctions.getConfig(configFile, 'activeLearning', 'sigmaN', sigmaN, 'float', True)
+
+    self.X = []
+    self.yBin = []  # .shape = (number of samples, number of unique classes)
+    self.yUni = []
+    self.w = []  # .shape = (feat dim, number of unique classes)
+
+
+  def checkModel(self):
+
+    if not numpy.all(numpy.isfinite(self.w)):
+      raise Exception('not numpy.all(numpy.isfinite(self.w))')
+
+    if not numpy.all(numpy.isfinite(self.X)):
+      raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+    if not numpy.all(numpy.isfinite(self.yUni)):
+      raise Exception('not numpy.all(numpy.isfinite(self.yUni))')
+
+
+  def train2(self, X, y, sigmaN=None):
+
+    self.train(X[0,:],y[0,:],sigmaN)
+    for idx in range(1,X.shape[0]):
+      self.update(X[idx,:],y[idx,:])
+
+
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y, sigmaN=None):
+
+    if sigmaN is not None:
+        self.sigmaN = sigmaN
+
+    # get all known classes
+    self.yUni = numpy.asmatrix(numpy.unique(numpy.asarray(y)))
+
+    # save stuff
+    self.X = X
+
+    # get binary labels for each ovr - classifier
+    self.yBin = numpy.asmatrix(numpy.empty((y.shape[0],self.yUni.shape[1])), dtype=numpy.int)
+    for cls in range(self.yUni.shape[1]):
+      self.yBin[:,cls] = 2*(y == self.yUni[0,cls]) - 1
+
+    # sample reweighting
+    rewDiagMat = helperFunctions.getReweightDiagMat(y, self.yUni)
+    rewX = rewDiagMat*self.X
+    rewY = rewDiagMat*self.yBin
+
+    # calculate w
+    #self.w = numpy.linalg.solve((rewX.T*rewX) + numpy.identity(X.shape[1])*self.sigmaN, rewX.T)*rewY
+    self.w = helperFunctions.solveW(rewX, rewY, self.sigmaN)
+
+    self.checkModel()
+
+
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+
+    # update samples
+    self.X = numpy.append(self.X, x, axis=0)
+
+    # update binary labels for known class
+    self.yBin = numpy.append(self.yBin, 2*(y == self.yUni) - 1, axis=0)
+
+    # create labels and w for new class
+    if not (self.yUni == y).any():
+
+        # get insertion idx
+        idx = numpy.searchsorted(numpy.ravel(numpy.asarray(self.yUni)), y[0,0])
+
+        # store new label and find new class index
+        self.yUni = numpy.insert(self.yUni, [idx], y, axis=1)
+
+        ## add new binary label vector for new class
+        self.yBin = numpy.insert(self.yBin, [idx], numpy.zeros((self.yBin.shape[0],1), dtype=numpy.int), axis=1)
+        self.yBin[:-1,idx] = -1
+        self.yBin[-1,idx] = 1
+
+        # sample reweighting
+        rewDiagMat = helperFunctions.getReweightDiagMat(helperFunctions.getYfromYbin(self.yBin, self.yUni), self.yUni)
+        rewX = rewDiagMat*self.X
+        rewY = rewDiagMat*self.yBin[:,idx]
+
+        # train new w for new class
+        #self.w = numpy.insert(self.w, [idx], numpy.linalg.solve((rewX.T*rewX) + numpy.identity(x.shape[1])*self.sigmaN, rewX.T)*rewY, axis=1)
+        self.w = numpy.insert(self.w, [idx], helperFunctions.solveW(rewX, rewY, self.sigmaN), axis=1)
+
+    # sample reweighting
+    rewDiagMat = helperFunctions.getReweightDiagMat(helperFunctions.getYfromYbin(self.yBin, self.yUni), self.yUni)
+    rewX = rewDiagMat*self.X
+    rewY = rewDiagMat*self.yBin
+
+    # update w
+    self.w = helperFunctions.solveW(rewX, rewY, self.sigmaN, self.w)
+
+    self.checkModel()
+
+
+  # X.shape = (number of samples, feat dim), loNoise = {0,1}
+  def infer(self, x, loNoise=False):
+
+    loNoise = loNoise and (self.yUni == -1).any()
+
+    pred =  numpy.asmatrix(numpy.dot(x,self.w[:,int(loNoise):]))
+
+    if not numpy.all(numpy.isfinite(pred)):
+      raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+    return pred
+
+
+  # X.shape = (number of samples, feat dim)
+  def test(self, x, loNoise=False):
+
+    loNoise = loNoise and (self.yUni == -1).any()
+
+    return self.yUni[0,numpy.argmax(self.infer(x, loNoise), axis=1) + int(loNoise)]
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcSigmaF(self, x, tmpVec=None, rewX=None):
+
+    if tmpVec is None:
+        if rewX is None:
+            rewDiagMat = helperFunctions.getReweightDiagMat(helperFunctions.getYfromYbin(self.yBin, self.yUni), self.yUni)
+            rewX = numpy.dot(rewDiagMat,self.X)
+
+        sigmaF = numpy.sum(numpy.multiply(x,(self.sigmaN*numpy.linalg.solve(numpy.add(numpy.dot(rewX.T,rewX), numpy.identity(x.shape[1])*self.sigmaN), x.T)).T),axis=1)
+    else:
+        sigmaF = self.sigmaN*tmpVec.T
+
+    if not numpy.all(numpy.isfinite(sigmaF)):
+        raise Exception('not numpy.all(numpy.isfinite(sigmaF))')
+
+    return sigmaF
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcProbs(self, x, mu=None, sigmaF=None, nmb=1000):
+
+    # get mu and sigma
+    if mu is None:
+      mu = self.infer(x,(self.yUni == -1).any())
+    if sigmaF is None:
+      sigmaF = self.calcSigmaF(x)
+
+    # prepare
+    probs = numpy.asmatrix(numpy.zeros(mu.shape))
+
+    for idx in range(nmb):
+
+      draws = numpy.asmatrix(numpy.random.randn(mu.shape[0],mu.shape[1]))
+      draws = numpy.add(numpy.multiply(draws, numpy.repeat(sigmaF, draws.shape[1], axis=1)), mu)
+
+      maxIdx = numpy.argmax(draws, axis=1)
+
+      idxs = (range(len(maxIdx)),numpy.squeeze(maxIdx))
+      probs[idxs] = probs[idxs] + 1
+
+    # convert absolute to relative amount
+    return probs/float(nmb)
+
+
+  # x.shape = (feat dim, number of samples)
+  def calcAlScores(self, x):
+
+    return None
+
+
+  # x.shape = (feat dim, number of samples)
+  def getAlScores(self, x):
+
+    alScores = self.calcAlScores(x)
+
+    if not numpy.all(numpy.isfinite(alScores)):
+      raise Exception('not numpy.all(numpy.isfinite(alScores))')
+
+    if alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1:
+      raise Exception('alScores.shape[0] != x.shape[0] or alScores.shape[1] != 1')
+
+    return alScores
+
+
+  # x.shape = (feat dim, number of samples)
+  def chooseSample(self, x):
+
+    return numpy.argmax(self.getAlScores(x), axis=0).item(0)

datasetAcquisition.py

@@ -0,0 +1,495 @@
+import scipy.io
+import numpy
+import pickle
+import os
+import time
+import sys
+
+import helperFunctions
+
+###
+
+#TODO: reuse readData functions for readDataForInit
+def readDataForInit(taskIdx, rndInitIdx, configFile):
+
+    readAttemptsNmb = helperFunctions.getConfig(configFile, 'data', 'readAttemptsNmb', 5, 'int', True)
+    readAttemptsDelay = helperFunctions.getConfig(configFile, 'data', 'readAttemptsDelay', 30, 'int', True)
+    datatype = helperFunctions.getConfig(configFile, 'data', 'datatype', None, 'str', True)
+    readAttempt = 0
+
+    while True:
+        try:
+
+            if datatype == 'mat':
+                xTrain, yTrain, xPool, yPool, xTest, yTest = readFromMatForInit(taskIdx, rndInitIdx, configFile)
+
+            elif  datatype == 'pickle':
+                xTrain, yTrain, xPool, yPool, xTest, yTest = readFromPickleForInit(taskIdx, rndInitIdx, configFile)
+
+            elif  datatype == 'usps':
+                xTrain, yTrain, xPool, yPool, xTest, yTest = readUSPSForInit(taskIdx, rndInitIdx, configFile)
+
+            elif  datatype == 'lfw':
+                xTrain, yTrain, xPool, yPool, xTest, yTest = readLFWForInit(taskIdx, rndInitIdx, configFile)
+
+            else:
+                raise Exception('Unknown datatype %s!'%datatype)
+
+            break
+
+        except:
+            readAttempt = readAttempt + 1
+            if readAttempt >= readAttemptsNmb:
+                raise Exception('ERROR: Reading datatype {} failed {} times!'.format(datatype, readAttempt))
+            print ''
+            print 'WARNING: Reading datatype {} failed ({} / {}, retry after {} seconds)!'.format(datatype, readAttempt, readAttemptsNmb, readAttemptsDelay)
+            sys.stdout.flush()
+            time.sleep(readAttemptsDelay)
+
+    ###
+
+    checkData(xTrain, yTrain, 'training')
+    checkData(xPool, yPool, 'pool')
+    checkData(xTest, yTest, 'test')
+
+    ###
+
+    return xTrain, yTrain, xPool, yPool, xTest, yTest
+
+
+###
+
+
+def checkData(x, y, identifier):
+
+    if x.shape[0] != y.shape[0]:
+        raise Exception('{} data: #x = {} != #y = {}'.format(identifier, x.shape[0],y.shape[0]))
+
+    if not numpy.all(numpy.isfinite(x)):
+        raise Exception('{} data: not numpy.all(numpy.isfinite(x))'.format(identifier))
+
+    if not numpy.all(numpy.isfinite(y)):
+        raise Exception('{} data: not numpy.all(numpy.isfinite(y))'.format(identifier))
+
+
+###
+
+
+def readFromMatForInit(taskIdx, rndInitIdx, configFile):
+
+    dataFilePattern = helperFunctions.getConfig(configFile, 'data', 'dataFilePattern', None, 'str', True)
+
+    mat = scipy.io.loadmat(dataFilePattern %(taskIdx + 1, rndInitIdx + 1))
+
+    xTrain = numpy.asmatrix(mat['xTrain'], dtype=numpy.double)
+    yTrain = numpy.asmatrix(mat['yTrain'], dtype=numpy.int)
+    xPool = numpy.asmatrix(mat['xPool'], dtype=numpy.double)
+    yPool = numpy.asmatrix(mat['yPool'], dtype=numpy.int)
+    xTest = numpy.asmatrix(mat['xTest'], dtype=numpy.double)
+    yTest = numpy.asmatrix(mat['yTest'], dtype=numpy.int)
+
+    return xTrain, yTrain, xPool, yPool, xTest, yTest
+
+
+###
+
+
+def readFromPickleForInit(taskIdx, rndInitIdx, configFile):
+
+    indicesFileName = helperFunctions.getConfig(configFile, 'data', 'indicesFileName', None, 'str', True)
+    trainFileName = helperFunctions.getConfig(configFile, 'data', 'trainFileName', None, 'str', True)
+    testFileName = helperFunctions.getConfig(configFile, 'data', 'testFileName', None, 'str', True)
+    noiseFileName = helperFunctions.getConfig(configFile, 'data', 'noiseFileName', None, 'str', True)
+    forbiddenCls = helperFunctions.getConfig(configFile, 'data', 'forbiddenCls', [], 'intList', True)
+    numNoiseSamples = helperFunctions.getConfig(configFile, 'data', 'numNoiseSamples', 20000, 'int', True)
+
+    ###
+
+    pickleIn = open(indicesFileName)
+    indices = pickle.load(pickleIn)
+    pickleIn.close()
+
+    trainIdxs = indices['trainIdxs'][taskIdx][rndInitIdx]
+    testIdxs = indices['testIdxs'][taskIdx][rndInitIdx]
+
+    ###
+
+    pickleIn = open(testFileName)
+    testData = pickle.load(pickleIn)
+    pickleIn.close()
+    yTest = numpy.asmatrix(testData['y'][testIdxs,:], dtype=numpy.int)
+    xTest = numpy.asmatrix(testData['X'][testIdxs,:], dtype=numpy.double)
+
+    del testData
+
+    ###
+
+    pickleIn = open(trainFileName)
+    trainData = pickle.load(pickleIn)
+    pickleIn.close()
+    yTrain = numpy.asmatrix(trainData['y'][trainIdxs,:], dtype=numpy.int)
+    xTrain = numpy.asmatrix(trainData['X'][trainIdxs,:], dtype=numpy.double)
+
+    ###
+
+    poolIdxs = numpy.delete(numpy.asarray(range(trainData['y'].shape[0])), trainIdxs)
+    yPool = numpy.asmatrix(trainData['y'][poolIdxs,:])
+    xPool = numpy.asmatrix(trainData['X'][poolIdxs,:])
+
+    del trainData
+
+    for fCLs in forbiddenCls:
+        idxs = numpy.where(numpy.asarray(yPool) == fCLs)
+        yPool = numpy.delete(yPool, idxs[0], axis=0)
+        xPool = numpy.delete(xPool, idxs[0], axis=0)
+
+    pickleIn = open(noiseFileName)
+    noiseData = pickle.load(pickleIn)
+    pickleIn.close()
+
+    yPool = numpy.append(yPool, numpy.asmatrix(noiseData['y'][:numNoiseSamples,:], dtype=numpy.int), axis=0)
+    xPool = numpy.append(xPool, numpy.asmatrix(noiseData['X'][:numNoiseSamples,:], dtype=numpy.double), axis=0)
+
+    del noiseData
+
+    ###
+
+    return xTrain, yTrain, xPool, yPool, xTest, yTest
+
+
+###
+
+
+def readUSPSForInit(taskIdx, rndInitIdx, configFile):
+
+    indicesFileName = helperFunctions.getConfig(configFile, 'data', 'indicesFileName', None, 'str', True)
+    trainFileName = helperFunctions.getConfig(configFile, 'data', 'trainFileName', None, 'str', True)
+    testFileName = helperFunctions.getConfig(configFile, 'data', 'testFileName', None, 'str', True)
+    noiseFileName = helperFunctions.getConfig(configFile, 'data', 'noiseFileName', None, 'str', True)
+    numNoiseSamples = helperFunctions.getConfig(configFile, 'data', 'numNoiseSamples', 1797, 'int', True)
+
+    ###
+
+    trainData = scipy.io.loadmat(trainFileName)['optdigits']
+    testData = scipy.io.loadmat(testFileName)['optdigits']
+    noiseData = scipy.io.loadmat(noiseFileName)['optdigits']
+
+    pickleIn = open(indicesFileName)
+    indices = pickle.load(pickleIn)
+    pickleIn.close()
+
+    trainIdxs = indices['trainIdxs'][taskIdx][rndInitIdx]
+    poolIdxs = numpy.delete(numpy.asarray(range(trainData.shape[0])), trainIdxs)
+    testIdxs = indices['testIdxs'][taskIdx][rndInitIdx]
+
+    ###
+
+    xTrain = numpy.asmatrix(trainData[trainIdxs,:-1], dtype=numpy.double)
+    yTrain = numpy.asmatrix(trainData[trainIdxs,-1], dtype=numpy.int).T
+
+    xPool = numpy.asmatrix(trainData[poolIdxs,:-1], dtype=numpy.double)
+    yPool = numpy.asmatrix(trainData[poolIdxs,-1], dtype=numpy.int).T
+
+    xPool = numpy.append(xPool, numpy.asmatrix(noiseData[:numNoiseSamples,:], dtype=numpy.double), axis=0)
+    yPool = numpy.append(yPool, numpy.asmatrix(numpy.ones((numNoiseSamples,1))*(-1.0), dtype=numpy.int), axis=0)
+
+    xTest = numpy.asmatrix(testData[:,:-1], dtype=numpy.double)
+    yTest = numpy.asmatrix(testData[:,-1], dtype=numpy.int).T
+
+    ###
+
+    return xTrain, yTrain, xPool, yPool, xTest, yTest
+
+
+###
+
+
+def readLFWForInit(taskIdx, rndInitIdx, configFile):
+
+    indicesFileName = helperFunctions.getConfig(configFile, 'data', 'indicesFileName', None, 'str', True)
+    dataFileName = helperFunctions.getConfig(configFile, 'data', 'dataFileName', None, 'str', True)
+    numMinSamples = helperFunctions.getConfig(configFile, 'data', 'numMinSamples', 55, 'int', True)
+    numNoiseSamples = helperFunctions.getConfig(configFile, 'data', 'numNoiseSamples', 400, 'int', True)
+    splitFileName =  helperFunctions.getConfig(configFile, 'data', 'splitFileName', os.path.join(os.path.dirname(dataFileName),'lfwRndTestSplit.pickle'), 'str', True)
+    numTestSplit = helperFunctions.getConfig(configFile, 'data', 'numTestSplit', 30, 'int', True)
+
+    ###
+
+    mat = scipy.io.loadmat(dataFileName)
+    labels = numpy.asmatrix(mat['labels'], dtype=numpy.int)
+    data = numpy.asmatrix(mat['data'], dtype=numpy.float)
+
+    if not os.path.isfile(splitFileName):
+
+        print 'creating new split file ...'
+
+        cls = numpy.unique(numpy.asarray(labels))
+        noiseCls = list()
+        noiseIdxs = numpy.empty((0,), dtype=int)
+        trainIdxs = numpy.empty((0,), dtype=int)
+        testIdxs = numpy.empty((0,), dtype=int)
+
+        for clsIdx in range(len(cls)):
+
+            clsIndices = numpy.ravel(numpy.asarray(numpy.where(numpy.ravel(numpy.asarray(labels)) == cls[clsIdx])))
+
+            if len(clsIndices) < numMinSamples:
+                noiseCls.append(cls[clsIdx])
+                noiseIdxs = numpy.append(noiseIdxs, clsIndices, axis=0)
+
+            else:
+
+                clsIndices = clsIndices[numpy.random.permutation(len(clsIndices))]
+                testIdxs = numpy.append(testIdxs, clsIndices[:numTestSplit], axis=0)
+                trainIdxs = numpy.append(trainIdxs, clsIndices[numTestSplit:], axis=0)
+
+        noiseIdxs = noiseIdxs[numpy.random.permutation(len(noiseIdxs))[:numNoiseSamples]]
+
+        outputFile = open(splitFileName, 'w')
+        pickle.dump({'noiseCls':noiseCls, 'noiseIdxs':noiseIdxs, 'trainIdxs':trainIdxs, 'testIdxs':testIdxs}, outputFile)
+        outputFile.close()
+
+    else:
+
+        print 'loading split file ...'
+
+        pickleIn = open(testFileName)
+        splitIndices = pickle.load(pickleIn)
+        pickleIn.close()
+
+        noiseIdxs = splitIndices['noiseIdxs']
+        trainIdxs = splitIndices['trainIdxs']
+        testIdxs = splitIndices['testIdxs']
+
+    ###
+
+    xData = data[trainIdxs,:]
+    yData = labels[trainIdxs,:]
+
+    xData = numpy.append(xData, data[noiseIdxs,:], axis=0)
+    yData = numpy.append(yData, numpy.asmatrix(numpy.ones((numNoiseSamples,1))*(-1.0), dtype=numpy.int), axis=0)
+
+    xTest = data[testIdxs,:]
+    yTest = labels[testIdxs,:]
+
+    ###
+
+    pickleIn = open(indicesFileName)
+    indices = pickle.load(pickleIn)
+    pickleIn.close()
+
+    curTrainIdxs = indices['trainIdxs'][taskIdx][rndInitIdx]
+    curPoolIdxs = numpy.delete(numpy.asarray(range(yData.shape[0])), trainIdxs)
+    curTestIdxs = indices['testIdxs'][taskIdx][rndInitIdx]
+
+    ###
+
+    xTrain = xData[curTrainIdxs,:]
+    yTrain = yData[curTrainIdxs,:]
+
+    xPool = xData[curPoolIdxs,:]
+    yPool = yData[curPoolIdxs,:]
+
+    xTest = xTest[curTestIdxs,:]
+    yTest = yTest[curTestIdxs,:]
+
+    ###
+
+    return xTrain, yTrain, xTest, yTest
+
+
+###
+
+
+def readData(configFile):
+
+    readAttemptsNmb = helperFunctions.getConfig(configFile, 'data', 'readAttemptsNmb', 5, 'int', True)
+    readAttemptsDelay = helperFunctions.getConfig(configFile, 'data', 'readAttemptsDelay', 30, 'int', True)
+    datatype = helperFunctions.getConfig(configFile, 'data', 'datatype', None, 'str', True)
+    readAttempt = 0
+
+    while True:
+        try:
+
+            if  datatype == 'pickle':
+                xTrain, yTrain, xTest, yTest = readFromPickle(configFile)
+
+            if  datatype == 'usps':
+                xTrain, yTrain, xTest, yTest = readUSPS(configFile)
+
+            if  datatype == 'lfw':
+                xTrain, yTrain, xTest, yTest = readLFW(configFile)
+
+            else:
+                raise Exception('Unknown datatype %s!'%datatype)
+
+            break
+
+        except:
+            readAttempt = readAttempt + 1
+            if readAttempt >= readAttemptsNmb:
+                raise Exception('ERROR: Reading datatype {} failed {} times!'.format(datatype, readAttempt))
+            print ''
+            print 'WARNING: Reading datatype {} failed ({} / {}, retry after {} seconds)!'.format(datatype, readAttempt, readAttemptsNmb, readAttemptsDelay)
+            sys.stdout.flush()
+            time.sleep(readAttemptsDelay)
+
+    ###
+
+    checkData(xTrain, yTrain, 'training')
+    checkData(xTest, yTest, 'test')
+
+    ###
+
+    return xTrain, yTrain, xTest, yTest
+
+
+###
+
+
+def readFromPickle(configFile):
+
+    trainFileName = helperFunctions.getConfig(configFile, 'data', 'trainFileName', None, 'str', True)
+    testFileName = helperFunctions.getConfig(configFile, 'data', 'testFileName', None, 'str', True)
+    noiseFileName = helperFunctions.getConfig(configFile, 'data', 'noiseFileName', None, 'str', True)
+    forbiddenCls = helperFunctions.getConfig(configFile, 'data', 'forbiddenCls', [], 'intList', True)
+    numNoiseSamples = helperFunctions.getConfig(configFile, 'data', 'numNoiseSamples', 20000, 'int', True)
+
+    ###
+
+    pickleIn = open(testFileName)
+    testData = pickle.load(pickleIn)
+    pickleIn.close()
+    yTest = numpy.asmatrix(testData['y'], dtype=numpy.int)
+    xTest = numpy.asmatrix(testData['X'], dtype=numpy.double)
+
+    del testData
+
+    ###
+
+    pickleIn = open(trainFileName)
+    trainData = pickle.load(pickleIn)
+    pickleIn.close()
+
+    pickleIn = open(noiseFileName)
+    noiseData = pickle.load(pickleIn)
+    pickleIn.close()
+
+    yTrain = numpy.append(trainData['y'], numpy.asmatrix(noiseData['y'][:numNoiseSamples,:], dtype=numpy.int), axis=0)
+    xTrain = numpy.append(trainData['X'], numpy.asmatrix(noiseData['X'][:numNoiseSamples,:], dtype=numpy.double), axis=0)
+
+    del trainData
+    del noiseData
+
+    ###
+
+    return xTrain, yTrain, xTest, yTest
+
+
+###
+
+
+def readUSPS(configFile):
+
+    trainFileName = helperFunctions.getConfig(configFile, 'data', 'trainFileName', None, 'str', True)
+    testFileName = helperFunctions.getConfig(configFile, 'data', 'testFileName', None, 'str', True)
+    noiseFileName = helperFunctions.getConfig(configFile, 'data', 'noiseFileName', None, 'str', True)
+    numNoiseSamples = helperFunctions.getConfig(configFile, 'data', 'numNoiseSamples', 1797, 'int', True)
+
+    ###
+
+    trainData = scipy.io.loadmat(trainFileName)['optdigits']
+    testData = scipy.io.loadmat(testFileName)['optdigits']
+    noiseData = scipy.io.loadmat(noiseFileName)['optdigits']
+
+    ###
+
+    xTrain = numpy.asmatrix(trainData[:,:-1], dtype=numpy.double)
+    yTrain = numpy.asmatrix(trainData[:,-1], dtype=numpy.int).T
+
+    xTrain = numpy.append(xTrain, numpy.asmatrix(noiseData[:numNoiseSamples,:], dtype=numpy.double), axis=0)
+    yTrain = numpy.append(yTrain, numpy.asmatrix(numpy.ones((numNoiseSamples,1))*(-1.0), dtype=numpy.int), axis=0)
+
+    xTest = numpy.asmatrix(testData[:,:-1], dtype=numpy.double)
+    yTest = numpy.asmatrix(testData[:,-1], dtype=numpy.int).T
+
+    ###
+
+    return xTrain, yTrain, xTest, yTest
+
+
+###
+
+
+def readLFW(configFile):
+
+    dataFileName = helperFunctions.getConfig(configFile, 'data', 'dataFileName', None, 'str', True)
+    numMinSamples = helperFunctions.getConfig(configFile, 'data', 'numMinSamples', 55, 'int', True)
+    numNoiseSamples = helperFunctions.getConfig(configFile, 'data', 'numNoiseSamples', 400, 'int', True)
+    splitFileName =  helperFunctions.getConfig(configFile, 'data', 'splitFileName', os.path.join(os.path.dirname(dataFileName),'lfwRndTestSplit.pickle'), 'str', True)
+    numTestSplit = helperFunctions.getConfig(configFile, 'data', 'numTestSplit', 30, 'int', True)
+
+    ###
+
+    mat = scipy.io.loadmat(dataFileName)
+    labels = numpy.asmatrix(mat['labels'], dtype=numpy.int)
+    data = numpy.asmatrix(mat['data'], dtype=numpy.float)
+
+    if not os.path.isfile(splitFileName):
+
+        print 'creating new split file ...'
+
+        cls = numpy.unique(numpy.asarray(labels))
+        noiseCls = list()
+        noiseIdxs = numpy.empty((0,), dtype=int)
+        trainIdxs = numpy.empty((0,), dtype=int)
+        testIdxs = numpy.empty((0,), dtype=int)
+
+        for clsIdx in range(len(cls)):
+
+            clsIndices = numpy.ravel(numpy.asarray(numpy.where(numpy.ravel(numpy.asarray(labels)) == cls[clsIdx])))
+
+            if len(clsIndices) < numMinSamples:
+                noiseCls.append(cls[clsIdx])
+                noiseIdxs = numpy.append(noiseIdxs, clsIndices, axis=0)
+
+            else:
+
+                clsIndices = clsIndices[numpy.random.permutation(len(clsIndices))]
+                testIdxs = numpy.append(testIdxs, clsIndices[:numTestSplit], axis=0)
+                trainIdxs = numpy.append(trainIdxs, clsIndices[numTestSplit:], axis=0)
+
+        noiseIdxs = noiseIdxs[numpy.random.permutation(len(noiseIdxs))[:numNoiseSamples]]
+
+        outputFile = open(splitFileName, 'w')
+        pickle.dump({'noiseCls':noiseCls, 'noiseIdxs':noiseIdxs, 'trainIdxs':trainIdxs, 'testIdxs':testIdxs}, outputFile)
+        outputFile.close()
+
+    else:
+
+        print 'loading split file ...'
+
+        pickleIn = open(testFileName)
+        splitIndices = pickle.load(pickleIn)
+        pickleIn.close()
+
+        noiseIdxs = splitIndices['noiseIdxs']
+        trainIdxs = splitIndices['trainIdxs']
+        testIdxs = splitIndices['testIdxs']
+
+    ###
+
+    xTrain = data[trainIdxs,:]
+    yTrain = labels[trainIdxs,:]
+
+    xTrain = numpy.append(xTrain, data[noiseIdxs,:], axis=0)
+    yTrain = numpy.append(yTrain, numpy.asmatrix(numpy.ones((numNoiseSamples,1))*(-1.0), dtype=numpy.int), axis=0)
+
+    xTest = data[testIdxs,:]
+    yTest = labels[testIdxs,:]
+
+    ###
+
+    return xTrain, yTrain, xTest, yTest
+
+

evaluation/PrecomputeExperimentalSetup.py

@@ -0,0 +1,153 @@
+
+import pickle
+import socket
+import datetime
+import numpy
+
+###
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+
+###
+
+import helperFunctions
+import datasetAcquisition
+
+###
+
+if len(sys.argv) != 2:
+    raise Exception('No config file given!')
+
+print ''
+print ' -- config -- '
+print ''
+
+numTasks = helperFunctions.getConfig(sys.argv[1], 'experiment', 'numTasks', None, 'int', True)
+numRndInits = helperFunctions.getConfig(sys.argv[1], 'experiment', 'numRndInits', None, 'int', True)
+numCls = helperFunctions.getConfig(sys.argv[1], 'experiment', 'numCls', None, 'int', True)
+numInitCls = helperFunctions.getConfig(sys.argv[1], 'experiment', 'numInitCls', None, 'int', True)
+numInitSamplesPerCls = helperFunctions.getConfig(sys.argv[1], 'experiment', 'numInitSamplesPerCls', None, 'int', True)
+numTestSamplesPerCls = helperFunctions.getConfig(sys.argv[1], 'experiment', 'numTestSamplesPerCls', None, 'int', True)
+
+forbiddenCls = helperFunctions.getConfig(sys.argv[1], 'data', 'forbiddenCls', [], 'intList', True)
+indicesFileName = helperFunctions.getConfig(configFile, 'data', 'indicesFileName', None, 'str', True)
+
+print ''
+print 'host:', socket.gethostname()
+print 'pid:', os.getpid()
+print 'now:', datetime.datetime.strftime(datetime.datetime.now(), '%d.%m.%Y %H:%M:%S')
+print 'git:', helperFunctions.getGitHash()
+print ''
+sys.stdout.flush()
+
+###
+
+trainIdxs = list()
+testIdxs = list()
+
+###
+
+print''
+xTrain, yTrain, xTest, yTest = datasetAcquisition.readData(sys.argv[1])
+
+###
+
+print ''
+print ' -- train --'
+print ''
+uniY = numpy.unique(numpy.asarray(yTrain))
+
+###
+
+for fCLs in forbiddenCls:
+    uniY = numpy.delete(uniY, numpy.where(uniY == fCLs), axis=0)
+
+for fCLs in forbiddenCls:
+    uniY = numpy.delete(uniY, numpy.where(uniY == -1), axis=0)
+
+###
+
+for taskIdx in range(numTasks):
+
+    initCls = uniY[numpy.random.permutation(len(uniY))[:numInitCls]]
+    taskList = list()
+
+    for rndInitIdx in range(numRndInits):
+
+        rndInitList = list()
+
+        for clsIdx in range(len(initCls)):
+
+            clsSamples = numpy.ravel(numpy.where(numpy.ravel(yTrain) == numpy.ravel(initCls[clsIdx])))
+
+            if rndInitIdx == 0:
+                print 'cls', initCls[clsIdx], 'with', len(clsSamples), 'samples chosen'
+                if len(clsSamples) < numInitSamplesPerCls:
+                    #raise Exception('To few samples!')
+                    print '>>> to few samples!'
+                sys.stdout.flush()
+
+            rndInitList.extend(list(clsSamples[numpy.random.permutation(len(clsSamples))[:numInitSamplesPerCls]]))
+
+        print 'samples gathered for rndinit', rndInitIdx, 'of task', taskIdx, ' -> ', len(rndInitList)
+
+        taskList.append(list(rndInitList))
+
+    trainIdxs.append(list(taskList))
+
+###
+
+print ''
+print ' -- test --'
+print ''
+
+uniY = numpy.unique(numpy.asarray(yTest))
+
+###
+
+for fCLs in forbiddenCls:
+    uniY = numpy.delete(uniY, numpy.where(uniY == fCLs), axis=0)
+
+for fCLs in forbiddenCls:
+    uniY = numpy.delete(uniY, numpy.where(uniY == -1), axis=0)
+
+###
+
+
+for taskIdx in range(numTasks):
+
+    taskList = list()
+
+    for rndInitIdx in range(numRndInits):
+
+        rndInitList = list()
+
+        for clsIdx in range(len(uniY)):
+
+            clsSamples = numpy.ravel(numpy.where(numpy.ravel(yTest) == numpy.ravel(uniY[clsIdx])))
+
+            if taskIdx == 0 and rndInitIdx == 0:
+                print 'samples available for cls', uniY[clsIdx], ' -> ', len(clsSamples)
+                if len(clsSamples) < numTestSamplesPerCls:
+                    #raise Exception('To few samples!')
+                    print '>>> to few samples!'
+                sys.stdout.flush()
+
+            rndInitList.extend(list(clsSamples[numpy.random.permutation(len(clsSamples))[:numTestSamplesPerCls]]))
+
+        print 'samples gathered for rndinit', rndInitIdx, 'of task', taskIdx, ' -> ', len(rndInitList)
+
+        taskList.append(list(rndInitList))
+
+    testIdxs.append(list(taskList))
+
+###
+
+out = open(indicesFileName, 'w')
+pickle.dump({'trainIdxs': trainIdxs, 'testIdxs': testIdxs}, out)
+out.close()
+print ''
+print 'done'
+

evaluation/RunExperiment.py

@@ -0,0 +1,225 @@
+#! /usr/bin/python
+
+import scipy.io
+import numpy
+import time
+import socket
+import datetime
+
+###
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+
+###
+
+import helperFunctions
+import methodSelection
+import datasetAcquisition
+
+###
+
+if len(sys.argv) != 2:
+    raise Exception('No config file given!')
+
+print ''
+print ' -- config I -- '
+print ''
+
+defaultFname =  os.path.join(os.path.dirname(sys.argv[1]),os.pardir,'setup.cfg')
+if not os.path.isfile(defaultFname):
+    defaultFname = sys.argv[1]
+
+setupFileName = helperFunctions.getConfig(sys.argv[1], 'experiment', 'setupFileName', defaultFname, 'str', True)
+numTasks = helperFunctions.getConfig(setupFileName, 'experiment', 'numTasks', 3, 'int', True)
+numRndInits = helperFunctions.getConfig(setupFileName, 'experiment', 'numRndInits', 3, 'int', True)
+numSteps = helperFunctions.getConfig(setupFileName, 'experiment', 'numSteps', 500, 'int', True)
+numCls = helperFunctions.getConfig(setupFileName, 'experiment', 'numCls', 80, 'int', True)
+forbiddenCls = helperFunctions.getConfig(setupFileName, 'experiment', 'forbiddenCls', [], 'intList', True)
+notificationPath = helperFunctions.getConfig(setupFileName, 'experiment', 'notificationPath', None, 'str', True)
+writeAttemptsNmb = helperFunctions.getConfig(setupFileName, 'experiment', 'writeAttemptsNmb', 5, 'int', True)
+writeAttemptsDelay = helperFunctions.getConfig(setupFileName, 'experiment', 'writeAttemptsDelay', 30, 'int', True)
+
+print ''
+print ' -- config II -- '
+print ''
+
+rejectNoise = helperFunctions.getConfig(sys.argv[1], 'experiment', 'rejectNoise', True, 'bool', True)
+continueExperiment = helperFunctions.getConfig(sys.argv[1], 'experiment', 'continueExperiment', False, 'bool', True)
+useApproximation = helperFunctions.getConfig(sys.argv[1], 'experiment', 'prepareApproximation', False, 'bool', True)
+alMethod = helperFunctions.getConfig(sys.argv[1], 'activeLearning', 'method', None, 'str', True)
+rewMethod = helperFunctions.getConfig(sys.argv[1], 'reweighting', 'method', 'None', 'str', True)
+startTaskIdx = helperFunctions.getConfig(sys.argv[1], 'experiment', 'startTaskIdx', 0, 'int', True)
+endTaskIdx = helperFunctions.getConfig(sys.argv[1], 'experiment', 'endTaskIdx', numTasks - 1, 'int', True)
+
+if (startTaskIdx != endTaskIdx) or (numTasks < 2):
+    resultsFileName = helperFunctions.getConfig(sys.argv[1], 'experiment', 'resultsFileName', os.getcwd() + '/results.mat', 'str', True)
+else:
+    resultsFileName = helperFunctions.getConfig(sys.argv[1], 'experiment', 'resultsFileName', os.getcwd() + '/results' + str(startTaskIdx) + '.mat', 'str', True)
+
+identifier = helperFunctions.getConfig(sys.argv[1], 'experiment', 'identifier', os.path.basename(os.path.dirname(resultsFileName)), 'str', True)
+
+print ''
+print 'host:', socket.gethostname()
+print 'pid:', os.getpid()
+print 'now:', datetime.datetime.strftime(datetime.datetime.now(), '%d.%m.%Y %H:%M:%S')
+print 'git:', helperFunctions.getGitHash()
+sys.stdout.flush()
+
+if identifier is None or resultsFileName is None or alMethod is None:
+    raise Exception('ERROR: Config incomplete!')
+
+if not os.path.isdir(os.path.dirname(os.path.abspath(resultsFileName))) or not os.path.exists(os.path.dirname(os.path.abspath(resultsFileName))):
+    raise Exception('ERROR: Results path does not exist!')
+
+if os.getcwd() != os.path.dirname(os.path.abspath(resultsFileName)):
+    print ''
+    print 'WARNING: current path != results path'
+    print 'current:', os.getcwd()
+    print 'rerults:', os.path.dirname(os.path.abspath(resultsFileName))
+
+###
+
+if continueExperiment and os.path.isfile(resultsFileName):
+
+    print ''
+    print 'loading previous results ...'
+    sys.stdout.flush()
+
+    tmp = scipy.io.loadmat(resultsFileName)['results']
+    values = tmp.item(0)
+    names = list(tmp.dtype.names)
+
+    confMats = numpy.asarray(values[names.index('confMats')], dtype=numpy.float)
+    queriedIdxs = numpy.asarray(values[names.index('queriedIdxs')], dtype=numpy.float)
+    #name = values[names.index('name')].item(0)
+    identifier = values[names.index('identifier')].item(0)
+    knownCls = numpy.asarray(values[names.index('knownCls')], dtype=numpy.float)
+    timeNeeded = numpy.asarray(values[names.index('timeNeeded')], dtype=numpy.float)
+    startTaskIdx = values[names.index('lastTaskIdx')].item(0)
+    startRndInitIdx = values[names.index('lastRndInitIdx')].item(0) + 1
+
+else:
+    queriedIdxs = numpy.zeros((numTasks,numRndInits,numSteps))
+    timeNeeded = numpy.zeros((numTasks,numRndInits,numSteps))
+    confMats = numpy.zeros((numTasks,numRndInits,numSteps + 1,numCls,numCls))
+    knownCls = numpy.zeros((numTasks,numRndInits,numSteps + 1))
+    startRndInitIdx = 0
+
+###
+
+#numpy.random.seed(int(time.time()*1000.0))
+timePast = 0
+totalRuns = (endTaskIdx + 1)*numRndInits*numSteps - startTaskIdx*numRndInits*numSteps - startRndInitIdx*numSteps
+queriesPast = 0
+
+for taskIdx in range(startTaskIdx, endTaskIdx + 1):
+    for rndInitIdx in range(startRndInitIdx, numRndInits):
+
+        if notificationPath is not None:
+            helperFunctions.writeNotification(notificationPath, 'status__' + identifier + '__' + socket.gethostname() + '__' + str(taskIdx) + '__' + str(rndInitIdx))
+
+        print''
+        print 'loading data ...'
+        xTrain, yTrain, xPool, yPool, xTest, yTest = datasetAcquisition.readDataForInit(taskIdx, rndInitIdx, setupFileName)
+
+        print''
+        print 'training models ...'
+        classifier = methodSelection.selectActiveLearning(alMethod, sys.argv[1])
+        classifier.train(xTrain, yTrain)
+        reweighter = methodSelection.selectReweighter(rewMethod, sys.argv[1])
+        reweighter.train(xTrain, yTrain)
+        sys.stdout.flush()
+
+        pred = classifier.test(xTest)
+
+        confMats[taskIdx,rndInitIdx,0,:,:] = helperFunctions.confusionMatrix(yTest, pred)
+        knownCls[taskIdx,rndInitIdx,0] = classifier.yUni.shape[1]
+
+        print ''
+        print 'next task:', taskIdx, ', next rndInit:', rndInitIdx
+        print 'xTrain: {}, yTrain: {} [#cls: {}], xPool: {}, yPool: {} [#cls: {}, #noise: {}], xTest: {}, yTest: {} [#cls: {}]'.format(xTrain.shape, yTrain.shape, len(numpy.unique(numpy.asarray(yTrain))), xPool.shape, yPool.shape, len(numpy.unique(numpy.asarray(yPool))), numpy.sum(yPool==-1), xTest.shape, yTest.shape, len(numpy.unique(numpy.asarray(yTest))))
+        print 'initial acc:', helperFunctions.getAvgAcc(confMats[taskIdx,rndInitIdx,0,:,:]), ', initial knownCls:', int(knownCls[taskIdx,rndInitIdx,0])
+        sys.stdout.flush()
+
+        orgIdxs = numpy.asmatrix(range(1,yPool.shape[0] + 1))
+
+        if useApproximation:
+            print 'prepare approximation ...'
+            print
+            sys.stdout.flush()
+            classifier.prepareApprox(xPool)
+
+        for step in range(numSteps):
+
+            t0 = time.time()
+
+            alScores1 = classifier.getAlScores(xPool)
+            alScores2 = reweighter.reweight(alScores1, xPool)
+            chosenIdx = numpy.argmax(alScores2, axis=0).item(0)
+
+            newX = xPool[chosenIdx,:]
+            newY = yPool[chosenIdx,:]
+
+            reweighter.update(newX, newY)
+
+            if not(rejectNoise and newY == -1):
+                classifier.update(newX, newY)
+                if newY == -1:
+                    print '-- updated with noise'
+            else:
+                print '-- noise drawn and rejected'
+
+            queriedIdxs[taskIdx,rndInitIdx,step] = orgIdxs[0,chosenIdx]
+
+            pred = classifier.test(xTest, True)
+            confMats[taskIdx,rndInitIdx,step + 1,:,:] = helperFunctions.confusionMatrix(yTest, pred)
+
+            knownCls[taskIdx,rndInitIdx,step + 1] = classifier.yUni.shape[1] - (classifier.yUni == -1).any()
+
+            xPool = numpy.delete(xPool, (chosenIdx), axis=0)
+            yPool = numpy.delete(yPool, (chosenIdx), axis=0)
+            orgIdxs = numpy.delete(orgIdxs, (chosenIdx), axis=1)
+
+            if useApproximation:
+                classifier.clearApprox(chosenIdx)
+
+            t1 = time.time()
+            timeNeeded[taskIdx,rndInitIdx,step] = (t1 - t0)
+
+            queriesPast = queriesPast + 1
+
+            timePast = timePast + float(t1 - t0)
+            timePerPass = timePast / float(queriesPast)
+            timeOva = timePerPass * totalRuns
+            estTimeLeft = timeOva - timePast
+
+            print queriesPast, '/', totalRuns, '- time past:', '%.3f'%(timePast/3600.0), 'h, avg. time per pass:', '%.3f'%timePerPass, 's, est. time left:', '%.3f'%(estTimeLeft/3600.0), 'h, est. time over all:', '%.3f'%(timeOva/3600.0), 'h'
+            print 'chosenIdx:', chosenIdx,  ', task:', taskIdx, ', rndInit:', rndInitIdx, ', step:', step, ', acc:', '%.5f'%helperFunctions.getAvgAcc(confMats[taskIdx,rndInitIdx,step + 1,:,:]), ', knownCls:', int(knownCls[taskIdx,rndInitIdx,step + 1])
+            sys.stdout.flush()
+
+        results = dict(confMats=confMats, queriedIdxs=queriedIdxs, name=identifier, identifier=identifier, knownCls=knownCls, timeNeeded=timeNeeded, lastTaskIdx=taskIdx, lastRndInitIdx=rndInitIdx)
+
+        writeAttempt = 0
+        while True:
+            try:
+                scipy.io.savemat(resultsFileName, dict(results=results))
+                break
+            except:
+                writeAttempt = writeAttempt + 1
+                if writeAttempt >= writeAttemptsNmb:
+                    raise Exception('ERROR: Writing file {} failed {} times!'.format(resultsFileName, writeAttempt))
+                print ''
+                print 'WARNING: Writing file {} failed ({} / {}, retry after {} seconds)!'.format(resultsFileName, writeAttempt, writeAttemptsNmb, writeAttemptsDelay)
+                sys.stdout.flush()
+                time.sleep(writeAttemptsDelay)
+
+
+    startRndInitIdx = 0
+
+if notificationPath is not None:
+    helperFunctions.writeNotification(notificationPath, 'status__' + identifier + '__' + socket.gethostname() + '__done')
+
+print 'done'
+print 'now', datetime.datetime.strftime(datetime.datetime.now(), '%d.%m.%Y %H:%M:%S')

example_setup.cfg

@@ -0,0 +1,51 @@
+[experiment]
+
+# selection of experimental setup
+numTasks: 10
+numRndInits: 10
+numSteps: 100
+numCls: 10
+
+rejectNoise: False
+continueExperiment: True
+useApproximation: False
+
+
+[data]
+
+# dataset, see data for possible choices
+datatype: pickle
+
+# generated by PrecomputeExperimentalSetup.py
+indicesFileName: None
+
+# path to training feature file
+trainFileName: None
+
+# path to test feature file
+testFileName: None
+
+# path to noise feature file
+noiseFileName: None
+
+# comma separated list of forbidden class ids
+forbiddenCls: None
+
+# path to training feature file = helperFunctions.getConfig(configFile, 'data', 'forbiddenCls', [], 'intList', True)
+numNoiseSamples: None
+
+
+[activeLearning]
+
+# config for al method, see methodSelection.py for possible choices
+method: GPLinKemoc
+sigmaN: 0.00178
+useDensity: True
+
+
+[reweighting]
+
+# config for reweighting method, see methodSelection.py for possible choices
+method: GPLinK
+sigmaN: 0.00178
+

helperFunctions.py

@@ -0,0 +1,168 @@
+import numpy
+import math
+import ConfigParser
+import os
+import sys
+import subprocess
+import scipy.sparse.linalg
+import time
+
+
+def confusionMatrix(expected, predicted):
+
+    #print expected.shape, predicted.shape
+    #print numpy.unique(numpy.asarray(expected))
+
+    yUni = numpy.asmatrix(numpy.unique(numpy.asarray(expected)))
+    confMat = numpy.asmatrix(numpy.zeros((yUni.shape[1], yUni.shape[1])))
+
+    for expc, pred in zip(expected, predicted):
+        confMat[numpy.where(yUni == expc)[1], numpy.where(yUni == pred)[1]] += 1
+
+    if numpy.sum(confMat) != max(predicted.shape):
+        print numpy.sum(confMat), '!=', max(predicted.shape), '(', predicted.shape, ')'
+        print 'cls expected: ', numpy.unique(numpy.asarray(expected))
+        print 'cls predicted:', numpy.unique(numpy.asarray(predicted))
+        raise Exception('# predicted cls > # expected cls')
+
+    return confMat
+
+
+def getAvgAcc(confMat):
+
+    return numpy.mean(numpy.diagonal(confMat)/numpy.sum(confMat, axis=1).T)
+
+
+def getConfig(pathtoConfig, section, option, default=None, dtype='str', verbose=False):
+
+    # set default
+    value = default
+    defaultUsed = True
+
+    # check if file is available
+    if pathtoConfig is not None and os.path.isfile(pathtoConfig):
+
+        # init
+        config = ConfigParser.ConfigParser()
+        configFile = open(pathtoConfig)
+        config.readfp(configFile)
+        configFile.close()
+
+        # check if section and option is available
+        if config.has_section(section) and config.has_option(section, option):
+
+            # get requested type
+            if dtype == 'str':
+                value = config.get(section, option)
+            elif dtype == 'int':
+                value = config.getint(section, option)
+            elif dtype == 'float':
+                value = config.getfloat(section, option)
+            elif dtype == 'bool':
+                value = config.getboolean(section, option)
+            elif dtype == 'strList':
+                value = config.get(section, option).split(',')
+            elif dtype == 'intList':
+                value = [int(entry) for entry in config.get(section, option).split(',')]
+            elif dtype == 'floatList':
+                value = [float(entry) for entry in config.get(section, option).split(',')]
+            elif dtype == 'boolList':
+                value = [bool(entry) for entry in config.get(section, option).split(',')]
+            else:
+                raise Exception('Unknown dtype!')
+
+            defaultUsed = False
+
+    # print config
+    if verbose:
+        aux = ''
+        if 'List' in dtype and len(value) > 0:
+            aux = '| entryDtype:' + str(type(value[0]))
+        print 'default:', defaultUsed, '| section:', section, '| option:', option, '| value:', value, '| dtype:', type(value), aux
+
+    # return
+    return value
+
+
+def getGitHash(gitPath=os.path.dirname(os.path.abspath(__file__))):
+    curDir = os.getcwd()
+    os.chdir(gitPath)
+    gitHash = subprocess.check_output(['git', 'rev-parse', '--short', 'HEAD'], stderr=subprocess.STDOUT).strip()
+    os.chdir(curDir)
+    return gitHash
+
+
+def getYfromYbin(yBin, yUni):
+
+    y = numpy.asmatrix(numpy.zeros((yBin.shape[0],1), dtype=numpy.int))
+    for idx in range(yBin.shape[0]):
+        y[idx,0] = int(yUni[0,numpy.ravel(yBin[idx,:] == 1)])
+    return y
+
+
+def solveW(x, y, sigmaN, initW=None, maxIter=None):
+
+    linOpX = scipy.sparse.linalg.aslinearoperator(x)
+    w = numpy.asmatrix(numpy.empty((x.shape[1], y.shape[1])))
+
+    def matvecFunc(curW):
+        return linOpX.rmatvec(linOpX.matvec(curW)) + sigmaN*curW
+
+    for clsIdx in range(y.shape[1]):
+
+        if initW is not None:
+            initWbin = numpy.ravel(numpy.asarray(initW[:,clsIdx])).T
+        else:
+            initWbin = None
+
+        linOpW = scipy.sparse.linalg.LinearOperator((x.shape[1], x.shape[1]), matvec=matvecFunc, dtype=x.dtype)
+
+        solvedWbin,info = scipy.sparse.linalg.cg(linOpW, linOpX.rmatvec(y[:, clsIdx]), x0=initWbin, maxiter=maxIter)
+
+        if info != 0 and maxIter is None:
+            print ''
+            print 'WARNING: cg not converged!'
+            print ''
+
+        w[:,clsIdx] = numpy.asmatrix(solvedWbin).T
+
+    return w
+
+
+def getClsWeights(y, yUni):
+
+    clsWeights = numpy.empty(yUni.shape[1])
+    for clsIdx in range(yUni.shape[1]):
+        clsWeights[clsIdx] = y.shape[0] / (float(yUni.shape[1])*numpy.argwhere(y==yUni[0,clsIdx]).shape[0])
+
+    return clsWeights
+
+
+def writeNotification(notificationPath, statusStr):
+
+    try:
+        open(os.path.join(notificationPath, statusStr), 'a').close()
+    except:
+        print ''
+        print 'ERROR: writing notification to {} failed!'.format(notificationPath)
+        sys.stdout.flush()
+
+
+def getReweightDiagMat(y, yUni=None, clsWeights=None):
+
+    if yUni is None:
+        yUni = numpy.asmatrix(numpy.unique(numpy.asarray(y)))
+
+    if clsWeights is None:
+        clsWeights = getClsWeights(y, yUni)
+
+    sampleWeights = clsWeights[numpy.searchsorted(numpy.ravel(numpy.asarray(yUni)), numpy.ravel(numpy.asarray(y)))]
+    sampleWeights = numpy.prod([sampleWeights], axis=0)
+
+    return numpy.asmatrix(numpy.diag(numpy.sqrt(sampleWeights*numpy.ones(y.shape[0]))))
+
+
+def showProgressBarTerminal(current, total, pre):
+
+    sys.stdout.write('\r%s %0.2f %%'%(pre,(float(current)/float(total))*100.0))
+    sys.stdout.flush()

methodSelection.py

@@ -0,0 +1,72 @@
+#! /usr/bin/python
+
+import sys
+import os
+
+import helperFunctions
+
+###
+
+def selectActiveLearning(method, configFile=None):
+
+  sys.path.append(os.path.abspath(os.path.dirname(__file__)) + '/activeLearning')
+
+  if (method == 'LinGP') or (method == 'LinGPemoc'):
+    import activeLearningLinGPemoc
+    return activeLearningLinGPemoc.Classifier(configFile=configFile)
+
+  elif (method == 'LinGPapprox') or (method == 'LinGPemocApprox'):
+    import activeLearningLinGPemocApprox
+    return activeLearningLinGPemocApprox.Classifier(configFile=configFile)
+
+  ###
+
+  elif (method == 'GPLinK') or (method == 'GPLinKemoc'):
+    import activeLearningGPLinKemoc
+    return activeLearningGPLinKemoc.Classifier(configFile=configFile)
+
+  ###
+
+  elif method == 'GPGenKemoc':
+    import activeLearningGPGenKemoc
+    return activeLearningGPGenKemoc.Classifier(configFile=configFile)
+
+  ###
+
+  elif (method == 'LinGPwali') or (method == 'WlinGP1vs2'):
+    import activeLearningWlinGP1vs2
+    return activeLearningWlinGP1vs2.Classifier(configFile=configFile)
+
+  ###
+
+  else:
+    raise Exception('Unknown method %s!'%method)
+
+###
+
+def selectReweighter(method, configFile=None):
+
+  sys.path.append(os.path.abspath(os.path.dirname(__file__)) + '/reweighting')
+
+  if method == 'None':
+    import reweightNone
+    return reweightNone.Reweighter(configFile=configFile)
+
+  elif method == 'GPLinK':
+    import reweightGPLinK
+    return reweightGPLinK.Reweighter(configFile=configFile)
+
+  elif method == 'GPGenK':
+    import reweightGPGenK
+    return reweightGPGenK.Reweighter(configFile=configFile)
+
+  elif method == 'LinGP':
+    import reweightLinGP
+    return reweightLinGP.Reweighter(configFile=configFile)
+
+  elif (method == 'LinGPwali') or (method == 'WlinGP'):
+    import reweightWlinGP
+    return reweightWlinGP.Reweighter(configFile=configFile)
+
+  else:
+    raise Exception('Unknown method %s!'%method)

reweighting/reweightGPGenK.py

@@ -0,0 +1,157 @@
+#! /usr/bin/python
+
+import numpy
+import scipy.special
+import sklearn.kernel_ridge
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class Reweighter:
+
+    def __init__(self, sigmaN=0.01, gamma=None, kernel='rbf', configFile=None):
+
+        self.sigmaN = helperFunctions.getConfig(configFile, 'activeLearning', 'sigmaN', sigmaN, 'float', True)
+        self.kernel = helperFunctions.getConfig(configFile, 'activeLearning', 'kernel', kernel, 'str', True)
+        self.gamma = helperFunctions.getConfig(configFile, 'activeLearning', 'gamma', gamma, 'float', True)
+        self.numKernelCores = helperFunctions.getConfig(configFile, 'activeLearning', 'numKernelCores', 1, 'int', True)
+
+        self.K = []
+        self.alpha = []
+        self.X = []
+        self.y = []
+
+        self.allowedKernels = ['rbf', 'sigmoid', 'polynomial', 'poly', 'linear', 'cosine']
+
+        if self.kernel not in self.allowedKernels:
+            raise Exception('Unknown kernel %s!'%self.kernel)
+
+
+    def kernelFunc(self, x1, x2=None, gamma=None):
+
+        if gamma is not None:
+            self.gamma = gamma
+
+        if self.kernel in ['rbf']:
+            return numpy.asmatrix(sklearn.kernel_ridge.pairwise_kernels(x1, x2, metric=self.kernel, gamma=self.gamma, n_jobs=self.numKernelCores), dtype=numpy.float)
+        else:
+            return numpy.asmatrix(sklearn.kernel_ridge.pairwise_kernels(x1, x2, metric=self.kernel, n_jobs=self.numKernelCores), dtype=numpy.float)
+
+
+    def checkModel(self):
+
+        if not numpy.all(numpy.isfinite(self.K)):
+            raise Exception('not numpy.all(numpy.isfinite(self.K))')
+
+        if not numpy.all(numpy.isfinite(self.alpha)):
+            raise Exception('not numpy.all(numpy.isfinite(self.alpha))')
+
+        if not numpy.all(numpy.isfinite(self.X)):
+            raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+        if not numpy.all(numpy.isfinite(self.y)):
+            raise Exception('not numpy.all(numpy.isfinite(self.y))')
+
+
+    # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+    def train(self, X, y):
+
+        self.X = X
+        self.y = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+        self.K = self.kernelFunc(X,X)
+        self.alpha = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float32))*self.sigmaN, self.y)
+
+        self.checkModel()
+
+
+    # x.shape = (1, feat dim), y.shape = (1, 1)
+    def update(self, x, y):
+
+        # relabel
+        relY = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+        # get new kernel columns
+        k =  self.kernelFunc(self.X, x)
+        selfK =  self.getSelfK(x)
+
+        # update alpha
+        term1 = 1.0 / (self.sigmaN + self.calcSigmaF(x, k, selfK).item(0))
+
+        term2 = numpy.asmatrix(numpy.ones((self.X.shape[0] + 1,x.shape[0])), dtype=numpy.float) * (-1.0)
+        term2[0:self.X.shape[0],:] = numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), k)
+
+        term3 = self.infer(x, k=k) - relY
+
+        self.alpha = numpy.add(numpy.append(self.alpha, numpy.zeros((1,self.alpha.shape[1])), axis=0), numpy.dot(numpy.dot(term1,term2),term3))
+
+        # update K
+        self.K = numpy.append(numpy.append(self.K, k, axis=1), numpy.append(k.T, selfK, axis=1), axis=0)
+
+        # update samples
+        self.X = numpy.append(self.X, x, axis=0)
+        self.y = numpy.append(self.y, y, axis=0)
+
+        self.checkModel()
+
+
+    # X.shape = (number of samples, feat dim), loNoise = {0,1}
+    def infer(self, x, loNoise=False, k=None):
+
+        if k is None:
+            k = self.kernelFunc(self.X, x)
+
+        loNoise = loNoise and (self.yUni == -1).any()
+
+        pred = numpy.asmatrix(numpy.dot(k.T,self.alpha[:,int(loNoise):]))
+
+        if not numpy.all(numpy.isfinite(pred)):
+            raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+        return pred
+
+
+    def calcSigmaF(self, x, k=None, selfK=None):
+
+        if k is None:
+            k = self.kernelFunc(self.X, x)
+
+        if selfK is None:
+            selfK = self.getSelfK(x)
+
+        sigmaF = numpy.subtract(selfK, numpy.sum(numpy.multiply(numpy.linalg.solve(numpy.add(self.K, numpy.identity(self.X.shape[0], dtype=numpy.float)*self.sigmaN), k),k).T, axis=1))
+
+        if not numpy.all(numpy.isfinite(sigmaF)):
+            raise Exception('not numpy.all(numpy.isfinite(sigmaF))')
+
+        return sigmaF
+
+
+    def getSelfK(self, x):
+
+        selfK = numpy.asmatrix(numpy.empty([x.shape[0],1], dtype=numpy.float))
+        for idx in range(x.shape[0]):
+            selfK[idx,:] = self.kernelFunc(x[idx,:])
+
+        return selfK
+
+
+    # X.shape = (number of samples, feat dim)
+    def reweight(self, alScores, x, k=None):
+
+        if k is None:
+            k = self.kernelFunc(self.X, x)
+
+        clsScores = self.infer(x, k=k)
+        var = self.calcSigmaF(x, k=k)
+
+        probs = 0.5 - 0.5*scipy.special.erf(clsScores*(-1)/numpy.sqrt(2*var))
+        newAlScores = numpy.multiply(alScores,(1 - probs))
+
+        if not numpy.all(numpy.isfinite(newAlScores)):
+            raise Exception('not numpy.all(numpy.isfinite(newAlScores))')
+
+        return newAlScores
+

reweighting/reweightGPLinK.py

@@ -0,0 +1,132 @@
+#! /usr/bin/python
+
+import numpy
+import scipy.special
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class Reweighter:
+
+  def __init__(self,
+               sigmaN = 0.00178,
+               configFile=None):
+
+    self.sigmaN = helperFunctions.getConfig(configFile, 'reweighting', 'sigmaN', sigmaN, 'float', True)
+
+    self.K = []
+    self.alpha = []
+    self.X = []
+    self.y = []
+
+
+  def checkModel(self):
+
+    if not numpy.all(numpy.isfinite(self.K)):
+      raise Exception('not numpy.all(numpy.isfinite(self.K))')
+
+    if not numpy.all(numpy.isfinite(self.alpha)):
+      raise Exception('not numpy.all(numpy.isfinite(self.alpha))')
+
+    if not numpy.all(numpy.isfinite(self.X)):
+      raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+    if not numpy.all(numpy.isfinite(self.y)):
+      raise Exception('not numpy.all(numpy.isfinite(self.y))')
+
+
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y):
+
+    self.X = X
+    self.y = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+    self.K = X*X.T
+    self.alpha = numpy.linalg.solve(self.K + numpy.identity(self.X.shape[0], dtype=numpy.float32)*self.sigmaN, self.y)
+
+    self.checkModel()
+
+
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+
+    # relabel
+    relY = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+    # get new kernel columns
+    k = self.X*x.T
+    selfK = numpy.sum(numpy.multiply(x,x), axis=1)
+
+    # get score of new sample
+    infY = self.infer(x, k=k)
+
+    # update alpha
+    term1 = 1.0 / (self.sigmaN + self.calcSigmaF(x, k, selfK).item(0))
+
+    term2 = numpy.asmatrix(numpy.ones((self.X.shape[0] + 1,x.shape[0])), dtype=numpy.float32) * -1.0
+    term2[0:self.X.shape[0],:] = numpy.linalg.solve(self.K + numpy.identity(self.X.shape[0], dtype=numpy.float32)*self.sigmaN, k)
+
+    term3 = infY - relY
+
+    self.alpha = numpy.append(self.alpha, numpy.zeros((1,self.alpha.shape[1])), axis=0) + term1*term2*term3
+
+    # update K
+    self.K = numpy.append(numpy.append(self.K, k, axis=1), numpy.append(k.T, selfK, axis=1), axis=0)
+
+    # update samples
+    self.X = numpy.append(self.X, x, axis=0)
+    self.y = numpy.append(self.y, relY, axis=0)
+
+    self.checkModel()
+
+
+  # X.shape = (number of samples, feat dim)
+  def infer(self, x, k=None):
+
+    if k is None:
+      k = self.X*x.T
+
+    pred = numpy.asmatrix(k.T*self.alpha)
+
+    if not numpy.all(numpy.isfinite(pred)):
+      raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+    return pred
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcSigmaF(self, x, k=None, selfK=None):
+
+    if k is None:
+        k = self.X*x.T
+
+    if selfK is None:
+        selfK = numpy.sum(numpy.multiply(x,x), axis=1)
+
+    sigmaF = selfK - numpy.sum(numpy.multiply(numpy.linalg.solve((self.K + numpy.identity(self.X.shape[0], dtype=numpy.float32)*self.sigmaN), k),k).T, axis=1)
+
+    if not numpy.all(numpy.isfinite(sigmaF)):
+      raise Exception('not numpy.all(numpy.isfinite(sigmaF))')
+
+    return sigmaF
+
+
+  # X.shape = (number of samples, feat dim)
+  def reweight(self, alScores, x, k=None):
+
+    if k is None:
+      k = self.X*x.T
+
+    clsScores = self.infer(x,k)
+    var = self.calcSigmaF(x,k)
+
+    probs = 0.5 - 0.5*scipy.special.erf(clsScores*(-1)/numpy.sqrt(2*var))
+    newAlScores = numpy.multiply(alScores,(1 - probs))
+
+    if not numpy.all(numpy.isfinite(newAlScores)):
+        raise Exception('not numpy.all(numpy.isfinite(newAlScores))')
+
+    return newAlScores
+

reweighting/reweightLinGP.py

@@ -0,0 +1,109 @@
+#! /usr/bin/python
+
+import numpy
+import scipy.special
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class Reweighter:
+
+  def __init__(self,
+                    sigmaN = 0.00178,
+                    configFile=None):
+
+    self.sigmaN = helperFunctions.getConfig(configFile, 'reweighting', 'sigmaN', sigmaN, 'float', True)
+
+    self.invCreg = []
+    self.w = []  # .shape = (feat dim, number of unique classes)
+    self.X = []
+    self.y = []
+
+
+  def checkModel(self):
+
+    if not numpy.all(numpy.isfinite(self.invCreg)):
+      raise Exception('not numpy.all(numpy.isfinite(self.invCreg))')
+
+    if not numpy.all(numpy.isfinite(self.w)):
+      raise Exception('not numpy.all(numpy.isfinite(self.w))')
+
+    if not numpy.all(numpy.isfinite(self.X)):
+      raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+    if not numpy.all(numpy.isfinite(self.y)):
+      raise Exception('not numpy.all(numpy.isfinite(self.y))')
+
+
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y):
+
+    # save stuff
+    self.X = X
+    self.y = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+    # calculate inverse of C_reg
+    self.invCreg = numpy.linalg.inv((self.X.T*self.X) + numpy.identity(self.X.shape[1])*self.sigmaN)
+
+    # calculate w
+    self.w = self.invCreg*self.X.T*self.y
+
+    self.checkModel()
+
+
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+
+    # relabel
+    relY = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+    # update w and C_reg
+    tmpVec = self.invCreg*x.T
+    tmpScalar = 1.0 + x*tmpVec
+    self.w = self.w + tmpVec*((relY - x*self.w)/tmpScalar)
+    self.invCreg = self.invCreg - ((tmpVec*tmpVec.T)/tmpScalar)
+
+    # update samples
+    self.X = numpy.append(self.X, x, axis=0)
+    self.y = numpy.append(self.y, relY, axis=0)
+
+    self.checkModel()
+
+
+  # X.shape = (number of samples, feat dim)
+  def infer(self, x):
+
+    pred = numpy.asmatrix(x*self.w)
+
+    if not numpy.all(numpy.isfinite(pred)):
+      raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+    return pred
+
+
+  # X.shape = (number of samples, feat dim)
+  def calcSigmaF(self, x):
+
+    sigmaF = numpy.sum(numpy.multiply(x,(self.sigmaN*self.invCreg*x.T).T),axis=1)
+
+    if not numpy.all(numpy.isfinite(sigmaF)):
+      raise Exception('not numpy.all(numpy.isfinite(sigmaF))')
+
+    return sigmaF
+
+
+  # X.shape = (number of samples, feat dim)
+  def reweight(self, alScores, x):
+
+    clsScores = self.infer(x)
+    var = self.calcSigmaF(x)
+
+    probs = 0.5 - 0.5*scipy.special.erf(clsScores*(-1)/numpy.sqrt(2*var))
+    newAlScores = numpy.multiply(alScores,(1 - probs))
+
+    if not numpy.all(numpy.isfinite(newAlScores)):
+        raise Exception('not numpy.all(numpy.isfinite(newAlScores))')
+
+    return newAlScores

reweighting/reweightNone.py

@@ -0,0 +1,45 @@
+#! /usr/bin/python
+
+import numpy
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class Reweighter:
+    
+  def __init__(self,configFile=None):
+   
+    return None  
+  
+  
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y):    
+   
+    return None  
+    
+    
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+   
+    return None      
+    
+    
+  # X.shape = (number of samples, feat dim)
+  def infer(self, x, k=None):
+  
+    return numpy.asmatrix(numpy.zeros(x.shape[0],1))
+
+
+  # X.shape = (number of samples, feat dim)
+  def reweight(self, alScores, x, k=None):
+  
+    return alScores
+  
+  
+  # X.shape = (number of samples, feat dim)
+  def calcSigmaF(self, x, k=None, selfK=None):
+    
+    return numpy.asmatrix(numpy.zeros(x.shape[0],1))
+ 

reweighting/reweightWlinGP.py

@@ -0,0 +1,102 @@
+#! /usr/bin/python
+
+import numpy
+import scipy.special
+
+import sys
+import os
+sys.path.append(os.path.join(os.path.abspath(os.path.dirname(__file__)),os.pardir))
+import helperFunctions
+
+class Reweighter:
+
+  def __init__(self,
+               sigmaN = 0.00178,
+               configFile=None):
+
+    self.sigmaN = helperFunctions.getConfig(configFile, 'reweighting', 'sigmaN', sigmaN, 'float', True)
+
+    self.w = []  # .shape = (feat dim, number of unique classes)
+    self.X = []
+    self.y = []
+
+
+  def checkModel(self):
+
+    if not numpy.all(numpy.isfinite(self.w)):
+      raise Exception('not numpy.all(numpy.isfinite(self.w))')
+
+    if not numpy.all(numpy.isfinite(self.X)):
+      raise Exception('not numpy.all(numpy.isfinite(self.X))')
+
+    if not numpy.all(numpy.isfinite(self.y)):
+      raise Exception('not numpy.all(numpy.isfinite(self.y))')
+
+
+  # X.shape = (number of samples, feat dim), y.shape = (number of samples, 1)
+  def train(self, X, y):
+
+    # save stuff
+    self.X = X
+    self.y = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+    # sample reweighting
+    rewDiagMat = helperFunctions.getReweightDiagMat(self.y, numpy.asmatrix(numpy.unique(numpy.asarray(self.y))))
+    rewX = numpy.dot(rewDiagMat,self.X)
+    rewY = numpy.dot(rewDiagMat,self.y)
+
+    # calculate w
+    self.w = helperFunctions.solveW(rewX, rewY, self.sigmaN)
+
+    self.checkModel()
+
+
+  # x.shape = (1, feat dim), y.shape = (1, 1)
+  def update(self, x, y):
+
+    # relabel
+    relY = numpy.asmatrix(y == -1, dtype=numpy.int)*2 - 1
+
+    # update samples
+    self.X = numpy.append(self.X, x, axis=0)
+    self.y = numpy.append(self.y, relY, axis=0)
+
+    # sample reweighting
+    rewDiagMat = helperFunctions.getReweightDiagMat(self.y, numpy.asmatrix(numpy.unique(numpy.asarray(self.y))))
+    rewX = numpy.dot(rewDiagMat,self.X)
+    rewY = numpy.dot(rewDiagMat,self.y)
+
+    # calculate w
+    self.w = helperFunctions.solveW(rewX, rewY, self.sigmaN, self.w)
+
+    self.checkModel()
+
+
+  # X.shape = (number of samples, feat dim)
+  def infer(self, x):
+
+    pred =  numpy.asmatrix(x*self.w)
+
+    if not numpy.all(numpy.isfinite(pred)):
+      raise Exception('not numpy.all(numpy.isfinite(pred))')
+
+    return pred
+
+
+  # X.shape = (number of samples, feat dim)
+  def reweight(self, alScores, x):
+
+    clsScores = self.infer(x)
+
+    # sample reweighting
+    rewDiagMat = helperFunctions.getReweightDiagMat(self.y, numpy.asmatrix(numpy.unique(numpy.asarray(self.y))))
+    rewX = numpy.dot(rewDiagMat,self.X)
+    var = numpy.sum(numpy.multiply(x,(self.sigmaN*numpy.linalg.solve(numpy.add(numpy.dot(rewX.T,rewX), numpy.identity(x.shape[1])*self.sigmaN), x.T)).T),axis=1)
+
+    probs = 0.5 - 0.5*scipy.special.erf(clsScores*(-1)/numpy.sqrt(2*var))
+    newAlScores = numpy.multiply(alScores,(1 - probs))
+
+    if not numpy.all(numpy.isfinite(newAlScores)):
+        raise Exception('not numpy.all(numpy.isfinite(newAlScores))')
+
+    return newAlScores