Skip to content

List of Test Scripts

Clark Wong edited this page Nov 21, 2018 · 2 revisions
Script Name Use
ai_battle Generates competition statistics of the specified AI models, could be used for benchmarking.
calculate_score Unit tests for the scoring rules.
fc_network A little experiment on using fully connected network for classification / regression
hand_predictor_stream Trains a neural network model to predict tiles held by a heuristic based opponent, using an infinite stream of matching data.
handpredictor_test Tests a "hand predictor" model.
hand_predictor Trains a neural network model to predict tiles held by a heuristic based opponent, using collected matching data.
mjdeepq Trains the deep Q model by letting it play against heuristics based players and players that take random moves.
mjedeepq Trains the extended deep Q model* by letting it play against heuristics based players and players that take random moves.
mjpg Trains the policy gradient model by letting it play against heuristic players and players that take random moves.
mjpgf This experiment trains the policy gradient model by forcing it to follow moves given by heuristic players.
rule_base_mcts Allows an user to play against the Monte Carlo Tree Search based model (C++ implementation) in console.
rule_base_naive Allows an user to play against the Monte Carlo Tree Search based model (Python implementation) in console.
rule_base_q Trains the simple Q learning model by letting it play against heuristics based players and players that take random moves.
simpledeepq A made up scenario: training a deepq model to escape a maze.
simplepg A made up scenario: training a policy gradient model to escape a maze.
simpleq A made up scenario: training a q learning model to escape a maze.
start_game Starts a game for 4 in command line console.
start_tggame Starts a game for 4 in command line console. An image of the board shows up whenever a human player needs to take a move.
validate_hand Unit test for hand validating.

Extended deep Q model: Extended in the sense that the action set includes different kinds of Chow and Pong, like Chow on Bamboo tiles, Chow on Dots tiles, etc.

Clone this wiki locally