pomdp (version 1.2.4)

value_function: Value Function

Description

Extracts the value function from a solved model. Extracts the alpha vectors describing the value function. This is similar to policy() which in addition returns the action prescribed by the solution.

Usage

value_function(model, drop = TRUE)
plot_value_function(
  model,
  projection = NULL,
  epoch = 1,
  ylim = NULL,
  legend = TRUE,
  col = NULL,
  lwd = 1,
  lty = 1,
  ylab = "Value",
  ...
)

Value

the function as a matrix with alpha vectors as rows.

Arguments

model

a solved POMDP or MDP.

drop

logical; drop the list for converged converged, epoch-independent value functions.

projection

Sample in a projected belief space. See projection() for details.

epoch

the value function of what epoch should be plotted? Use 1 for converged policies.

ylim

the y limits of the plot.

legend

logical; show the actions in the visualization?

col

potting colors.

lwd

line width.

lty

line type.

ylab

label for the y-axis.

...

additional arguments are passed on to stats::line() or graphics::barplot()`.

Author

Michael Hahsler

Details

Plots the value function of a POMDP solution as a line plot. The solution is projected on two states (i.e., the belief for the other states is held constant at zero). The value function can also be visualized using plot_belief_space().

See Also

Other policy: estimate_belief_for_nodes(), optimal_action(), plot_belief_space(), plot_policy_graph(), policy(), policy_graph(), projection(), reward(), solve_POMDP(), solve_SARSOP()

Other POMDP: MDP2POMDP, POMDP(), accessors, actions(), add_policy(), plot_belief_space(), projection(), reachable_and_absorbing, regret(), sample_belief_space(), simulate_POMDP(), solve_POMDP(), solve_SARSOP(), transition_graph(), update_belief(), write_POMDP()

Other MDP: MDP(), MDP2POMDP, MDP_policy_functions, accessors, actions(), add_policy(), gridworld, reachable_and_absorbing, regret(), simulate_MDP(), solve_MDP(), transition_graph()

Examples

data("Tiger")
sol <- solve_POMDP(Tiger)
sol

# value function for the converged solution
value_function(sol)

plot_value_function(sol, ylim = c(0,20))

## finite-horizon problem
sol <- solve_POMDP(model = Tiger, horizon = 3, discount = 1,
  method = "enum")
sol

# inspect the value function for all epochs
value_function(sol)

plot_value_function(sol, epoch = 1, ylim = c(-5, 25))
plot_value_function(sol, epoch = 2, ylim = c(-5, 25))
plot_value_function(sol, epoch = 3, ylim = c(-5, 25))

if (FALSE) {
# using ggplot2 to plot the value function for epoch 3
library(ggplot2)
pol <- policy(sol)
ggplot(pol[[3]]) +
 geom_segment(aes(x = 0, y = `tiger-left`, xend = 1, yend = `tiger-right`, color = action)) +
 coord_cartesian(ylim = c(-5, 15)) + ylab("Value") + xlab("Belief space")
}