A utility function that transforms the actions, i.e., A list of integers into card matrix. Here we remove the six entries that are always

A utility function to obtain one-hot endoding

A utility function to encode the historical moves. We encode the historical 15 actions. If there is no 15 actions, we pad the features wi

A utility function to encode the number of bombs into one-hot representation.

A utility function encoding historical moves. We encode 15 moves. If there is no 15 moves, we pad with zeros.

This function obtains observations with imperfect information from the infoset. It has three branches since we encode different features

This function is called in the end of each game. It returns either 1/-1 for win/loss, or ADP, i.e., every bomb will double th

This function is called in the end of each game. It returns either 1/-1 for win/loss, or ADP, i.e., every bomb will double th

A utility function to process observations and move them to CUDA.

Every time reset is called, the environment will be re-initialized with a new deck of cards. This function is usually called

Convert a list of integers to the tensor representation See Figure 2 in https://arxiv.org/pdf/2106.06135.pdf

Obttain the landlord features. See Table 4 in https://arxiv.org/pdf/2106.06135.pdf

Obttain the landlord_down features. See Table 5 in https://arxiv.org/pdf/2106.06135.pdf

Obttain the landlord_up features. See Table 5 in https://arxiv.org/pdf/2106.06135.pdf

Simply return the action that is set previously.

Get optimal combinations of cards in hand

Create three optimizers for the three positions

This function will sample a batch from the buffers based on the indices received from the full queue. It will also free the indices by se

Performs a learning (optimization) step.

The environment uses this function to tell the dummy agent what to do.

This is the main funtion for training. It will first initilize everything, such as buffers, optimizers, etc. Then it will start subproces

Objective is wp/adp/logadp. It indicates whether considers bomb in reward calculation. Here, we use dummy agents. This is bec

Initialzie this environment wrapper