coopihc.inference.GoalInferenceWithUserPolicyGiven.GoalInferenceWithUserPolicyGiven

class GoalInferenceWithUserPolicyGiven(*args, user_policy_model=None, **kwargs)

Bases: coopihc.inference.BaseInferenceEngine.BaseInferenceEngine

An Inference Engine used by an assistant to infer the ‘goal’ of a user via Bayesian updating. The likelihood is given through an ELLDiscretePolicy, which acts as the model of the user behavior.

• Expectations of the engine

  This inference engine expects:

  • the agent to have a ‘belief’ state, corresponding to the discrete belief pmf. in its internal state:

  • the modeler to attach a user policy model to the engine inference_engine._attach_policy(user_policy_model)

  • the modeler to attach the set of possible user goals inference_engine.attach_set_theta(set_theta)

• Inference

  Bayesian updating in the discrete case. Computes for each target :math:` heta` the associated posterior probability, given an observation \(x\) and the last user action \(y\):

  \[P(\Theta = \theta | X=x, Y=y) = \frac{p(Y = y | \Theta = \theta, X=x)}{\sum_{\Theta} p(Y=y|\Theta = \theta, X=x)} P(\Theta = \theta).\]

  The likelihood model (directly obtainable from an ELLDiscretePolicy) \(p(Y = y | \Theta = \theta, X=x)\) has to be supplied

  # Define the likelihood model for the user policy
  # user_policy_model = XXX
  # assert isinstance(user_policy_model, ELLDiscretePolicy)
  
  inference_engine = GoalInferenceWithUserPolicyGiven()
  # Attach it to the engine
  inference_engine._attach_policy(user_policy_model)
  

  It also expects that the set of :math:` heta`’s is supplied:

  set_theta = [
      {
          ("user_state", "goal"): StateElement(
              t,
              discrete_space(numpy.array(<XXX>),
          )
      }
      for t in <potential_targets>
  ]
  
  inference_engine.attach_set_theta(set_theta)
  

• Render

  —- plot mode:

  This engine will plot a bar chart of the beliefs on the assistant axis

  —- text mode:

  This engine will print out the beliefs to the terminal

• Example files

  coopihczoo.pointing.assistants

Parameters

user_policy_model (EELDiscretePolicy, optional) – a model of the user policy, defaults to None

Methods

add_observation	add observation
attach_set_theta
bind	Bind function to the engine with a given name.
default_value	Apply this decorator to use self.agent_observation as default value to infer from if agent_observation = None
infer
render
reset	reset _summary_

Attributes

action	The agent's last action
host
observation	The last observation.
role
state	The current agent state
unwrapped

property action

The agent’s last action

Returns

agent action

Return type

State

add_observation(observation)

add observation

Add an observation to a buffer. If the buffer does not exist, create a naive buffer. The buffer has a size given by buffer length

Parameters

observation (State) – observation produced by an engine

bind(func, as_name=None)

Bind function to the engine with a given name. If as_name is None, then the func name is used.

Parameters

• func (function) – function to bind

• as_name (string, optional) – name of resulting method, defaults to None

Returns

bound method

Return type

method

default_value()

Apply this decorator to use self.agent_observation as default value to infer from if agent_observation = None

property observation

The last observation.

Returns

last observation

Return type

State

reset(random=True)

reset _summary_

Empty the buffer

Parameters

random (bool, optional) – whether to randomize parameters internal to the inference engine. This is provided in case of subclass the BaseInferenceEngine, defaults to True.

property state

The current agent state

Returns

agent state

Return type

State