Dependencies and Requirements

• python=3.6.7

• pandas=0.25.1

• matplotlib=2.2.2

• scipy=1.3.1

• numpy=1.17.0

• seaborn=0.8.1

• scikit-learn=0.21.2

• pytorch=1.2.0

• tensorboardx=1.6

• luigi=2.7.5

• tqdm=4.33

• requests=2.18.4

• jupyterlab=1.0.2

• ipywidgets=7.5.1

• diskcache=3.0.6

• pyspark=2.4.3

• psutil=5.2.2

• category_encoders

• plotly=4.4.1

• imbalanced-learn==0.4.3

• torchbearer==0.5.1

• pytorch-nlp==0.4.1

• unidecode==1.1.1

• streamlit==0.52.2

• gym==0.15.4

Free software: MIT license

pip install mars-gym

pip install https://github.com/deeplearningbrasil/mars-gym/archive/master.zip

tox

set PYTEST_ADDOPTS=--cov-append
tox

PYTEST_ADDOPTS=--cov-append tox

Usage

Simulate Example

mars-gym run interaction --project PROJECT \
--n-factors N_FACTORS --learning-rate LR --optimizer OPTIMIZER \
--epochs EPOCHS --obs-batch-size OBS_BATCH \
--batch-size BATCH_SIZE --num-episodes NUM_EP \
--bandit-policy BANDIT --bandit-policy-params BANDIT_PARAMS

Evaluate Example

mars-gym evaluate iteraction \
 --model-task-id MODEL_TASK_ID --fairness-columns "[]" \
 --direct-estimator-class DE_CLASS

Evaluation Module

mars-gym viz

Cite

Please cite the associated paper for this work if you use this code:

@misc{santana2020marsgym,
      title={MARS-Gym: A Gym framework to model, train, and evaluate Recommender Systems for Marketplaces},
      author={Marlesson R. O. Santana and Luckeciano C. Melo and Fernando H. F. Camargo and Bruno Brandão and Anderson Soares and Renan M. Oliveira and Sandor Caetano},
      year={2020},
      eprint={2010.07035},
      archivePrefix={arXiv},
      primaryClass={cs.IR}
}

0.0.1 (2020-06-27)

• First release on PyPI.