Create a new environment with python 3.10
conda create -n ocl_survey python=3.10
conda activate ocl_survey
- To not lose sanity:
conda install mamba
- Then, follow the steps in this order
mamba install matplotlib
pip install torch torchvision
mamba env update -f environment.yaml
- Add to your python path the ocl_survey directory and the avalanche directory
conda env config vars set PYTHONPATH=/home/.../ocl_survey:/home/ocl_survey/avalanche.git
- Add a deploy config in the config/deploy folder, precising results and dataset path
- Test the environment by launching main.py
cd experiments/
python main.py
The code is structured as follows:
- src/ # contains source code for the experiments
- factories/ # Contains the code to create models, strategies, and benchmarks. Most code additions should be done here
- method_factory.py
- model_factory.py
- benchmark_factory.py
- toolkit/ # Contains some utils functions, parallel evaluation plugins, modified strategies (hyperparameter addition) etc...
- config/ # Config directory used by hydra,
- config.yaml # Default config for normal experiments
- hp_config.yaml # Default config for hp selection
- results.yaml # The main results directory, defaults to ../results, you can change this
- strategy/ # Contains strategy-specific config files (one per strategy)
- optimizer/ # Contains optimizer-specific config files (one per optimizer type)
- evaluation/ # Contains evaluation config files (no evaluation, non parallel evaluation, parallel evaluation)
- benchmarks/ # Contains benchmark relative config (one per benchmark)
- experiments/ # Contains the overrides for given experiments (one per benchmark), i.e model, batch size etc..
- deploy/ # Folder to precise results and dataset path
- scheduler/ # Contains learning rate scheduling relative args (one per scheduler)
- experiments/
- main.py # Main entry point for every experiments, no modifications should be needed in this
- main_hp_tuning.py # Main file for the hyperparameter tuning, change here the options for hp search depending on the method
- scripts/ # Contains shell scripts for running i.e multiple seeds, linear probing
- tests/ # Some tests for special functionalities, some more should be added maybe more related to the experiments
To launch an experiment, start from the default config file and change the part that needs to change
python main.py strategy=er_ace experiment=split_cifar100 evaluation=parallel
It's also possible to override more fine-grained arguments
python main.py strategy=er_ace experiment=split_cifar100 evaluation=parallel strategy.alpha=0.7 optimizer.lr=0.05
Before running the script, you can display the full config with "-c job" option
python main.py strategy=er_ace experiment=split_cifar100 evaluation=parallel -c job
Results will be saved in the directory specified in results.yaml. Under the following structure:
<results_dir>/<strategy_name>_<benchmark_name>/<seed>/
Modify the strategy specific search parameters, search range etc ... inside main_hp_tuning.py then run
python main_hp_tuning.py strategy=er_ace experiment=split_cifar100
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent