GeniusAgent

The GeniusAgent class us a wrapper around the fastAPI calls used to communicate to and from the Genius agent. This class has the following capabilities enables by the API calls:

• POST /graph of genius model loaded from a JSON or from the GeniusModel class

• GET /graph of genius model and print/view its contents

• DELETE /graph to unload the model from the agent

• POST /infer to perform inference given some evidence and a variable of interest

• POST /learn to perform parameter learning given an input CSV or list of variables and their observations

• POST /actionselection to perform action selection given a POMDP model structure and observation vector

• POST /import to perform structure learning given a CSV data file

• POST /validate-csv to perform validation on a CSV data file

• POST /simulate to perform simulation on a POMDP model structure

• POST /auto-bin-one-off to perform automatic binning of a given column by clustering

To import the GeniusAgent class use: from genius_client_sdk.agent import GeniusAgent

Usage patterns

GeniusAgent(agent_http_protocol, agent_hostname, agent_port, auth_config)

Attributes

Variable	Type	Description
agent_http_protocol	str	The HTTP protocol of the agent to connect to.
agent_hostname	str	The hostname of the agent to connect to.
agent_port	str	The port of the agent to connect to.
agent_url	str	The URL of the agent to connect to. Concatenation of the http protocol, hostname, and port.
auth_config	AuthConfig	The auth configuration for the agent.
model	GeniusModel	Local storage of the model to be sent to the agent.

Parameters

Class initialization parameters.

Variable	Type	Default	Description
agent_http_protocol	str	http	The HTTP protocol of the agent to connect to.
agent_hostname	str	localhost	The hostname of the agent to connect to.
agent_port	str	3000	The port of the agent to connect to.
auth_config	AuthConfig	NoAuthConfig	The auth configuration for the agent.

Methods

act

Performs action selection using active inference given an observation and policy length.

The observation must be an integer corresponding to a possible observation category. If there are multiple observation modalities then the observation will be a list with an integer corresponding to each observation modality.

Parameter	Type	Default	Description
observation	Union[int, str, Dict[str, Union[int, str]]]	Required	The observation to send to the agent.
policy_len	int	2	The length of the policy.
verbose	bool	True	If true, prints out the action, inferred state, and policy distribution.
polling_timeout_sec	float	120	he maximum time to wait in seconds for interference to complete.
polling_frequency_sec	float	0.1	The time to wait in seconds in between checking if the task has completed.
learn_likelihoods	bool	False
learn_transitions	bool	False
learn_initial_state_priors	bool	False

Returns:

Type	Description
dict	A dictionary containing the results of action selection. See the action data reference for a comprehensive overview of the components of this data.

Example (single modality):

Assumes a JSON model file already exists.

agent = GeniusAgent()
agent.load_genius_model(model=model)

action_result = agent.act(observation=1, policy_len=2)

Example (multiple modalities):

Assumes a JSON model file already exists.

agent = GeniusAgent()
agent.load_genius_model(model=model)

action_result = agent.act(observation=[1,0,4], policy_len=2)

infer

Run inference based on target variable and evidence of other variables.

Parameter	Type	Default	Description
variables	Union[str, list]	Required	The name of the target variable to infer.
evidence	dict	2	A dictionary containing observed data.
verbose	bool	True	If true, prints the probabilities associated with the inference result.
polling_timeout_sec	float	120	The maximum time to wait in seconds for interference to complete.
polling_frequency_sec	float	0.1	The time to wait in seconds in between checking if the task has completed.

Returns:

Type	Description
dict	A dictionary containing the results of inference. This includes the input evidence as well as the inferred probability distribution over the target variable.

Example:

Assumes a JSON model file already exists.

agent = GeniusAgent()
agent.load_genius_model(model=model)

variables = "wet_grass"
evidence = {"rain": "yes", "sprinkler": "off"}

inference_result = agent.infer(variables=variables, evidence=evidence)

learn

Run learning based on observations of variables. Accepts either a list of variables and nested list of observations or a path to a CSV file.

• If inputing a list of variables and observations then the variables and observations arguments are required.

• If inputing a CSV file with data, then the csv_path variable is required only.

Parameter	Type	Default	Description
variables	list	None	A list of variables to learn.
observations	list	None	A nested list of observations for the variables.
csv_path	str	None	The path to a CSV file containing the observations.
verbose	bool	False	If True, prints out the JSON model file.
polling_timeout_sec	float	120	The maximum time to wait in seconds for interference to complete.
polling_frequency_sec	float	0.1	The time to wait in seconds in between checking if the task has completed.

Returns:

Type	Description
dict	A dictionary containing the results of learning. This includes a dict representation of the newly learned model file.

Example (from CSV):

agent = GeniusAgent()
learning_result = agent.learn(csv_path="sprinkler_data.csv")

Example (from Python):

agent = GeniusAgent()

variables = ["cloudy", "rain", "sprinkler", "wet_grass"]
observations = [["yes", "yes", "on", "yes"], ["no", "no", "off", "no"]]

learning_result = agent.learn(variables=variables, observations=observations)

load_genius_model

Sends an instance of the GeniusModel class to the agent.

Parameter	Type	Default	Description
model	GeniusModel	Required	A instance of the GeniusModel class.

Example:

agent = GeniusAgent()
agent.load_genius_model(model=model)

load_model_from_json

Sends a JSON representation of a model to the agent.

Parameter	Type	Default	Description
json_path	str	Required	A path to a JSON file containing the model.

Example:

agent = GeniusAgent()
agent.load_model_from_json(json_path="sprinkler.json")

get_model_from_server

Gets the JSON model from the remote agent and updates the local model.

Parameter	Type	Default	Description
verbose	bool	False	If True, prints out the JSON model.

log_model

Prints the JSON model to logs. Optionally, summarizes the model in a slightly more readable format.

Parameter	Type	Default	Description
summary	bool	False	If true, prints a summarized view of the model.
logging_level	int	20	The logging level to use.

unload_model

Unloads the model locally and from the server

Parameter	Type	Default	Description
-	-	-	-

validate_csv

Performs validation on a given CSV file.

Parameter	Type	Default	Description
csv_path	str	Required	The path to the CSV file.
polling_timeout_sec	float	120	The maximum time to wait in seconds for interference to complete.
polling_frequency_sec	float	0.1	The time to wait in seconds in between checking if the task has completed.

simulate

Performs a simulation on the model given a library, number of steps, policy length, and initial state.

Parameter	Type	Default	Description
num_steps	int	Required	The number of steps to simulate.
policy_len	int	Required	The length of the policy.
init_state	dict	Required	The initial state for the simulation.
env	dict	Optional	The environment for the simulation.
polling_timeout_sec	float	120	The maximum time to wait in seconds for interference to complete.
polling_frequency_sec	float	0.1	he time to wait in seconds in between checking if the task has completed. Defaults to 0.1 seconds.

Returns:

Type	Description
dict	A dictionary containing the simulation result.

auto_bin_one_off

Automatically bins a given column by clustering.

Parameter	Type	Default	Description
csv_path	str	Required	The path to the CSV file.
variables	dict[str, AutoBinVariable]	Required	A dictionary of variables to bin. The keys are the variable names and the values are AutoBinVariable instances.
polling_timeout_sec	float	120	The maximum time to wait in seconds for interference to complete.
polling_frequency_sec	float	0.1	The time to wait in seconds in between checking if the task has completed.

Returns:

Type	Description
dict	A dictionary containing the binning result.