This repository will contain a modified and extended version of the avalanche framework which allow for sequence wise evaluation, class imbalanced rehearsal set selection, comprehensive logging, strategic memory filling. The commits made to the repository are designed such that they help understand the changes made to the original package.

Julia Boehlke 552b40ff23 forgot update to introductory jupyter notebook 3 年之前
avalanche 3682c6c83c rhearsal learning and limited memory filling togehter with updates to introductory notebook 3 年之前
data d0db47a081 explaining the data and data stream creating notebook 3 年之前
scripts 552b40ff23 forgot update to introductory jupyter notebook 3 年之前
.gitignore 46218e7ac2 remove ignored files & environment setup description 3 年之前
README.md 4e30a9f43a comprehensive metrics and explantions in intro notebook 3 年之前

README.md

This Repository contains a folder called avalanche which is basically a fork from the avalanche framework at the 18.8.2021 (checkout bcba08cf4c) by ContinualAI found here:

https://github.com/ContinualAI/avalanche

Further the camera trap data and the way it was handled is described. The corresponding jupyter notbook is found in scripts/jupyter-notebooks

After cloning the package, create a new envionrment with python 3.8:

conda create --name nameofenvironement python=3.8

Then activate the environment conda activate nameofenvironement

Next, navigate the avalanche folder containing the setup.py script. Use the following commad to get dependencies and install avalanche the environment in developer mode, meaning that changes made in the avalanche folder are seen by the environment: pip install -e .

Next, you need to install pytorch, torchvision and the appropriate cudatoolkit. Check your CUDA Version with nvidia-smi and then get the command here:

https://pytorch.org/get-started/locally/ (conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch)

Lastly, use this command to use the envionrment in a jupyter notebook:

python -m ipykernel install --user --name=nameofenvironement