Welcome to the code repository of the paper "Gaussian Siwtchsampling: A Second-Order Approach to Active Learning".
First, create a virtual environment, and install poetry
python3 -m venv .venv/
. .venv/bin/activate
pip install -U pip
pip install poetry
Second, run poetry to install dependencies
poetry install
If you run into an error similar to "Failed to unlock collection", I suggest the solution here.
Third, export the python path
export PYTHONPATH=${PYTHONPATH}:${PWD}
ALl configurations are managed in the example_config.toml file. Here, you can set different strategies, query sizes, etc. under the active_learning bracket. To see the available strategies, check out the init.py file under activelearning/qustrategies/. For hyperparameters related to training (e.g. learning rate) are under the classification bracket.
To run an active learning experiment run
python3 training/classification/run.py --config example_config.toml
To train a network outside of an active learning setting (with the full training set) run
python3 training/classification/run.py --config example_config.toml
If you find our code/paper insightful please consider citing us!