Skip to contents

Available Actions in a State

Source: R/available_actions.R
available_actions.Rd

Determine the set of actions available in a state.

Usage

available_actions(model, state, neg_inf_reward = TRUE, stay_in_place = FALSE)

Arguments

model

a MDP object.

state

a character vector specifying the states.

neg_inf_reward

logical; consider an action that produced -Inf reward to all end states unavailable?

stay_in_place

logical; consider an action that results in the same state with a probability of 1 as unavailable. Note that this means that absorbing states have no available action!

Value

a character vector with the available actions.

a vector with the available actions.

Details

Unavailable actions are modeled as actions that have an immediate reward of -Inf in the reward function. For a maze, also actions that do not change the state can be considered unavailable.

See also

Other MDP: MDP(), absorbing_states(), act(), action_state_helpers, find_reachable_states(), reachable_states(), sample_MDP(), sample_MDP.MDPSample(), start, transition_graph(), transition_matrix(), unreachable_states()

Author

Michael Hahsler

Examples

data(DynaMaze)
gw_plot(DynaMaze)


# The following actions are always available:
DynaMaze$actions
#> [1] "up"    "right" "down"  "left" 

# only right and down is unavailable for s(1,1) because they
#   make the agent stay in place.
available_actions(DynaMaze, state = "s(1,1)", stay_in_place = TRUE)
#>           up right down  left
#> s(1,1) FALSE  TRUE TRUE FALSE

# An action that leaves the grid currently is allowed but does not do
# anything.
act(DynaMaze, "s(1,1)", "up")
#> $reward
#> [1] -1
#> 
#> $state_prime
#> [1] s(1,1)
#> 47 Levels: s(1,1) s(2,1) s(3,1) s(4,1) s(5,1) s(6,1) s(1,2) s(2,2) ... s(6,9)
#>