refine_plan.models.dbn_option
Subclass of Option for options represented using dynamic Bayesian networks.
Author: Charlie Street Owner: Charlie Street
Module Contents
Classes
A subclass of options represented using dynamic Bayesian networks. |
- class refine_plan.models.dbn_option.DBNOption(name, transition_dbn_path, reward_dbn_path, sf_list, enabled_cond, prune_dists=True)
Bases:
refine_plan.models.option.OptionA subclass of options represented using dynamic Bayesian networks.
DBNs are passed in as files using the .bifxml format used by pyAgrum. The DBNs assume all state factor values are strings. This is to stop lookup errors with integers. States can be passed in without considering this however. The transition function uses up to two variables for each state factor sf: sf0 and sft. The reward function should just use one variable per state factor, and one more called r which represents the reward. Some state factors may not appear in the transition or reward DBNs.
- Same as superclass, plus
- _transition_dbn
The transition function DBN.
- _reward_dbn
The reward function DBN.
- _sf_list
The list of state factors that make up the state space
- _enabled_cond
A Condition which captures the states where this option is enabled
- get_transition_prob(state, next_state)
Return the transition probability for an (s,s’) pair.
- Parameters:
state – The first state
next_state – The next state
- Returns:
The transition probability
- get_reward(state)
Return the reward for executing an option in a given state.
- Parameters:
state – The state we want the reward for
- Returns:
The reward at the state
- get_transition_prism_string()
Return a PRISM string which captures all transitions for this option.
- Returns:
The transition PRISM string
- get_reward_prism_string()
Return a PRISM string which captures all rewards for this option.
- Returns:
The reward PRISM string