pip install -r requirements.txt
Any agent in the agents directory can be tested:
python run.py --agents [Agent1 Agent2 ...] --metrics [Metric1 Metric2 ...] --batch [batchsize]
e.g. python run.py --agents 'RandomAgent(epochs=10)' 'GreedyAgent(epochs=10)' --metrics 'Regret(0.2)' --batch 100 --reps 10
--env 'GenericEnv("data/toy/20mer.csv")'
: use X, Y data in provided file.
--reps [N]
: average multiple trials.
--cutoff [N]
: limit to N batches.
--pretrain
: use pretraining data.
--nocorr
: compute no prediction correlations.
Install OpenAI gym:
pip install -e gym_batgirl
To use:
env = gym.make('batgirl-v0')
actions = env.reset('protein-BRCA1', batch=100)
obs, regret, done, info = env.step(actions[:100])