Dataset

You can generate, for every Chef\`s Hat simulation, a dataset that contains all the actions taken by all the agents playing the game.

The gane dataset is a great resource for studying agent behavior, extracting game statistics and much more.

Using the ‘save_dataset’ parameter on any game room, the dataset is saved on the ‘log_directory/Dataset’ directory. It will be saved in two formats: an easily redable .csv file, and a easily processable .pkl pandas Dataframe.

Dataset Structure

The game dataset has the following columns:

Dataset Columns

Parameter	Description
index	It saves the date on which the action happened.
round_number	The number of the round that the action happened. Everytime someone declares a pizza, a new round is computed.
agent_names	The name and order of the agents thaat are currently playing the game.
source	Indicates if it was an agent, or the system, who did the action
action_type	Displays the action type: START_EXPERIMENT, NEW_MATCH, DEAL_CARDS, DISCARD, DECLARE_PIZZA, END_MATCH, CARD_EXCHANGE, DECLARE_SPECIAL_ACTION, END_EXPERIMENT
action_description	Gives more information for each action. If it was a discard action, it follows the standard action representation notation: C11;Q1;J0 = 1x card value 11, with 0 jokers.
player_finished	Indicates if the player that did this action finished all the cards at his hand.
player_hands	Shows the current cards at every player hands, following the format [player1[], player2[], player3[], player4[]]
board_before	The board before the player made his action
board_after	The board after the player made his action
possible_actions	Display all the possible actions that player could do, following the standard action representation notation.
current_roles	Display the current roles of each player in the format [player1, player2, player3, player4].
match_score	Display the score of this match in the format [player1, player2, player3, player4].
game_score	Display the cumulative score of the entire game in the format [player1, player2, player3, player4].
game_performance_score	Display the cumulative game performance of the entire game in the format [player1, player2, player3, player4].

Every row of the dataset represents one action during the game. And for each action, there are specific information that are stored on the dataset rows.

The action flow within the dataset is as follows:

action_type = "NEW_EXPERIMENT"# Information saved in this row: match_number, game_score, current_roles

while (not stop_criteria):
    action_type ="NEW_MATCH" # Information saved in this row: current_roles, game_score
    action_type ="DEAL_CARDS" # Information saved in this row:  match_number, player_hands
    action_type ="DECLARE_SPECIAL_ACTION"# (only if someone declare it). match_number, source, current_roles, action_description (which special action was declared)
    action_type = "CARD_EXCHANGE"# (only after the first match). Information saved in this row: match_number, action_description (who gave cards to who), and updated players_hand


    while players not finished:
      action_type = "DISCARD"# Information saved in this row: match_numner, game_number, source (author of the discard), action description (the discarded cards/pass), player hands, board before, board after, possible actions, and if this player has finished his cards or not
      if pizza:
        action_type= "PIZZA" # Information saved in this row: match_number, round_numner, source(author of the pizza)


    action_type= "END_MATCH" #Information saved in this row: match_score, current_roles

action_type = "END_EXPERIMENT"#Information saved in this row: current_roles, game_score, performance_score