The released model of the paper 'Automatic Bridge Bidding by Deep Reinforcement Learning' in ECAI 2016 https://arxiv.org/pdf/1607.03290v1.pdf

The current release includes 4 trained models when the limited maximum bids are 2 to 5, and the dataset.

The dataset is saved in data_train.mat, cost_train.mat, data_validate.mat, cost_validate.mat, data_test.mat, cost_test.mat.

The dataset is generated by modifying the Code of Ho-Chun Yen, the author of 'Contract Bridge Bidding by Learning' http://www.csie.ntu.edu.tw/~htlin/paper/doc/wscpii15bridgebid.pdf

To run the cost-calculating code for the model, simply run the 'automatic_bridge_bidding_model.m' code in matlab.

The totalbid variable in automatic_bridge_bidding_model.m can be changed from 2 to 5, where a respective trained model will be loaded.

The release of the code is to allow comparison with automatic_bridge_bidding_with_deep_reinforcement_learning in the non-competing setting.

The model can be read by numpy and load into a python file if preferred.

Experiments for bidding against the current model can be achieved by bidding PASS for the current model when the cost of all possible bids are > 0.20.