private static GamblersWorldAction FindBestAction(
GamblersWorld world,
GamblersWorldState state,
GamblersValueTable valueTable,
IGamblersWorldRewarder rewarder)
{
var maxActionValue = double.MinValue;
var maxAction = new GamblersWorldAction(0);
foreach (var action in world.AvailableActions(state))
{
var actionValue = 0.0;
foreach (var(nextState, pNextState) in world.PossibleStates(state, action))
{
var nextStateValue = valueTable.Value(nextState);
var reward = rewarder.Reward(state, nextState, action);
actionValue += pNextState * (reward + nextStateValue);
}
if (actionValue > maxActionValue)
{
maxActionValue = actionValue;
maxAction = action;
}
}
return(maxAction);
}
private static void PrintPolicyActions(
GamblersWorld world,
IDeterministicPolicy <GamblersWorldState, GamblersWorldAction> policy)
{
foreach (var state in world.AllStates())
{
Console.WriteLine($"{state}: {policy.Action(state)}");
}
}
private static void PrintAllValues(
GamblersWorld world,
ValueTable <GamblersWorldState, GamblersWorldAction> values)
{
foreach (var state in world.AllStates())
{
Console.WriteLine($"{state}: {values.Value(state)}");
}
}
public static GreedyGamblersPolicy Create(
GamblersWorld world,
GamblersValueTable valueTable,
IGamblersWorldRewarder rewarder)
{
var greedyPolicy = new GreedyGamblersPolicy(world);
foreach (var state in world.AllStates())
{
var bestAction = FindBestAction(world, state, valueTable, rewarder);
greedyPolicy._actions[state.DollarsInHand] = bestAction.Stake;
}
return(greedyPolicy);
}
private static void UseDpToFindOptimalPolicy()
{
const double probabilityOfHeads = 0.4;
const int dollarsToWin = 100;
const int evaluationSweepsPerPolicyUpdate = 1;
var world = new GamblersWorld(probabilityOfHeads, dollarsToWin);
var rewarder = new GamblersWorldRewarder(world);
Console.WriteLine("Policy optimiser:");
var(policy, values) = DpPolicyOptimiser
.FindOptimalPolicy(world, rewarder, evaluationSweepsPerPolicyUpdate);
Console.WriteLine("Optimal policy values:");
PrintAllValues(world, values);
Console.WriteLine("Optimal policy stakes:");
PrintPolicyActions(world, policy);
}
private static void PlayGamesWithPolicies()
{
var probabilityOfHeads = 0.2;
int dollarsToWin = 100;
var world = new GamblersWorld(probabilityOfHeads, dollarsToWin);
// Console.WriteLine("random policy:");
// PlayGamesWithPolicy(world, new UniformRandomGamblersPolicy());
// Console.WriteLine("random policy:");
// PlayGamesWithPolicy(world, new AlwaysStake1DollarPolicy());
Console.WriteLine("always stake max policy:");
PlayGamesWithPolicy(world, new AlwaysStakeMaxPolicy(world));
// var (optimalPolicy, _) = FindOptimalPolicy(world, new GamblersWorldRewarder(world));
// Console.WriteLine("optimal policy:");
// PlayGamesWithPolicy(world, optimalPolicy);
}
private static void EvaluatePolicyValues()
{
const double probabilityOfHeads = 0.4;
const int dollarsToWin = 100;
var world = new GamblersWorld(probabilityOfHeads, dollarsToWin);
var rewarder = new GamblersWorldRewarder(world);
Console.WriteLine("Optimal policy:");
var(policy, values) = DpPolicyOptimiser.FindOptimalPolicy(world, rewarder);
var evalValues = new ValueTable <GamblersWorldState, GamblersWorldAction>(world);
evalValues.Evaluate(policy, rewarder);
PrintPolicyActions(world, policy);
PrintAllValues(world, values);
// Console.WriteLine("Always stake max policy:");
// var policy = new AlwaysStakeMaxPolicy(world);
// PrintPolicyActionsAndValues(world, policy, rewarder);
}
private static void PlayGamesWithPolicy(GamblersWorld world, IGamblersPolicy policy)
{
var player = new GamblersProblemPlayer(world, policy);
player.Play();
}
public GamblersValueTable(GamblersWorld world)
{
_world = world;
_values = new double[_world.AllStates().Count()];
_values[_world.DollarsToWin] = 1.0;
}
public GamblersProblemPlayer(GamblersWorld world, IGamblersPolicy policy)
{
_world = world;
_policy = policy;
}
public GamblersWorldRewarder(GamblersWorld world)
{
_world = world;
}
public AlwaysStakeMaxPolicy(GamblersWorld world)
{
_world = world;
}
private GreedyGamblersPolicy(GamblersWorld world)
{
_actions = new int[world.AllStates().Count()];
}
