rlcard.models¶

rlcard.models.bridge_rule_models¶

File name: models/bridge_rule_models.py Author: William Hale Date created: 11/27/2021

Bridge rule models

class rlcard.models.bridge_rule_models.BridgeDefenderNoviceRuleAgent¶

Bases: object

Agent always passes during bidding

eval_step(state)¶

Predict the action given the current state for evaluation.: Since the agents is not trained, this function is equivalent to step function.

Parameters:: state (numpy.array) – an numpy array that represents the current state
Returns:: the action_id predicted by the agent probabilities (list): The list of action probabilities
Return type:: action_id (int)

static step(state) → int¶

Predict the action given the current state.

Defender Novice strategy:

Case during make call:: Always choose PassAction.
Case during play card:: Choose a random action.

Parameters:: state (numpy.array) – an numpy array that represents the current state
Returns:: the action_id predicted
Return type:: action_id (int)

rlcard.models.doudizhu_rule_models¶

Dou Dizhu rule models

class rlcard.models.doudizhu_rule_models.DouDizhuRuleAgentV1¶

Bases: object

Dou Dizhu Rule agent version 1

static card_str2list(hand)¶

combine_cards(hand)¶: Get optimal combinations of cards in hand

eval_step(state)¶: Step for evaluation. The same to step

static list2card_str(hand_list)¶

static pick_chain(hand_list, count)¶

step(state)¶

Predict the action given raw state. A naive rule. :param state: Raw state from the game :type state: dict

Returns:: Predicted action
Return type:: action (str)

class rlcard.models.doudizhu_rule_models.DouDizhuRuleModelV1¶

Bases: Model

Dou Dizhu Rule Model version 1

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

rlcard.models.gin_rummy_rule_models¶

File name: models/gin_rummy_rule_models.py Author: William Hale Date created: 2/12/2020

Gin Rummy rule models

class rlcard.models.gin_rummy_rule_models.GinRummyNoviceRuleAgent¶

Bases: object

Agent always discards highest deadwood value card

eval_step(state)¶

Predict the action given the current state for evaluation.: Since the agents is not trained, this function is equivalent to step function.

Parameters:: state (numpy.array) – an numpy array that represents the current state
Returns:: the action predicted by the agent probabilities (list): The list of action probabilities
Return type:: action (int)

static step(state)¶

Predict the action given the current state.

Novice strategy:

Case where can gin:: Choose one of the gin actions.
Case where can knock:: Choose one of the knock actions.
Case where can discard:: Gin if can. Knock if can. Otherwise, put aside cards in some best meld cluster. Choose one of the remaining cards with highest deadwood value. Discard that card.
Case otherwise:: Choose a random action.

Parameters:: state (numpy.array) – an numpy array that represents the current state
Returns:: the action predicted
Return type:: action (int)

class rlcard.models.gin_rummy_rule_models.GinRummyNoviceRuleModel¶

Bases: Model

Gin Rummy Rule Model

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

rlcard.models.leducholdem_rule_models¶

Leduc Hold ‘em rule model

class rlcard.models.leducholdem_rule_models.LeducHoldemRuleAgentV1¶

Bases: object

Leduc Hold ‘em Rule agent version 1

eval_step(state)¶: Step for evaluation. The same to step

static step(state)¶

Predict the action when given raw state. A simple rule-based AI. :param state: Raw state from the game :type state: dict

Returns:: Predicted action
Return type:: action (str)

class rlcard.models.leducholdem_rule_models.LeducHoldemRuleAgentV2¶

Bases: object

Leduc Hold ‘em Rule agent version 2

eval_step(state)¶: Step for evaluation. The same to step

static step(state)¶

Predict the action when given raw state. A simple rule-based AI. :param state: Raw state from the game :type state: dict

Returns:: Predicted action
Return type:: action (str)

class rlcard.models.leducholdem_rule_models.LeducHoldemRuleModelV1¶

Bases: Model

Leduc holdem Rule Model version 1

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

class rlcard.models.leducholdem_rule_models.LeducHoldemRuleModelV2¶

Bases: Model

Leduc holdem Rule Model version 2

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

rlcard.models.limitholdem_rule_models¶

Limit Hold ‘em rule model

class rlcard.models.limitholdem_rule_models.LimitholdemRuleAgentV1¶

Bases: object

Limit Hold ‘em Rule agent version 1

eval_step(state)¶: Step for evaluation. The same to step

static step(state)¶

Predict the action when given raw state. A simple rule-based AI. :param state: Raw state from the game :type state: dict

Returns:: Predicted action
Return type:: action (str)

class rlcard.models.limitholdem_rule_models.LimitholdemRuleModelV1¶

Bases: Model

Limitholdem Rule Model version 1

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

property use_raw¶

Indicate whether use raw state and action

Returns:: True if using raw state and action
Return type:: use_raw (boolean)

rlcard.models.uno_rule_models¶

UNO rule models

class rlcard.models.uno_rule_models.UNORuleAgentV1¶

Bases: object

UNO Rule agent version 1

static count_colors(hand)¶

Count the number of cards in each color in hand

Parameters:: hand (list) – A list of UNO card string
Returns:: The number cards of each color
Return type:: color_nums (dict)

eval_step(state)¶: Step for evaluation. The same to step

static filter_wild(hand)¶

Filter the wild cards. If all are wild cards, we do not filter

Parameters:: hand (list) – A list of UNO card string
Returns:: A filtered list of UNO string
Return type:: filtered_hand (list)

step(state)¶

Predict the action given raw state. A naive rule. Choose the color: that appears least in the hand from legal actions. Try to keep wild cards as long as it can.

Parameters:: state (dict) – Raw state from the game
Returns:: Predicted action
Return type:: action (str)

class rlcard.models.uno_rule_models.UNORuleModelV1¶

Bases: Model

UNO Rule Model version 1

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

property use_raw¶

Indicate whether use raw state and action

Returns:: True if using raw state and action
Return type:: use_raw (boolean)

rlcard.models.pretrained_models¶

Wrrapers of pretrained models.

class rlcard.models.pretrained_models.LeducHoldemCFRModel¶

Bases: Model

A pretrained model on Leduc Holdem with CFR (chance sampling)

property agents¶

Get a list of agents for each position in a the game

Returns:: A list of agents
Return type:: agents (list)

Note: Each agent should be just like RL agent with step and eval_step: functioning well.

rlcard.models.registration¶

class rlcard.models.registration.ModelRegistry¶

Bases: object

Register a model by ID

load(model_id)¶

Create a model instance

Parameters:: model_id (string) – the name of the model

register(model_id, entry_point)¶

Register an model

Parameters:

model_id (string) – the name of the model
entry_point (string) – a string the indicates the location of the model class

class rlcard.models.registration.ModelSpec(model_id, entry_point=None)¶

Bases: object

A specification for a particular Model.

load()¶

Instantiates an instance of the model

Returns:: an instance of the Model
Return type:: Model (Model)

rlcard.models.registration.load(model_id)¶

Create and model instance

Parameters:: model_id (string) – the name of the model

rlcard.models.registration.register(model_id, entry_point)¶

Register a model

Parameters:

model_id (string) – the name of the model
entry_point (string) – a string the indicates the location of the model class