public static APolicyOptimizer getOptimizedPolicyInput(ParallelOptions parallelOptInner,
Dictionary <string, string> processArgs,
AGameGraph graph,
AGameProcess game,
IGameParams gameParams)
{
// use optimizer, if specified
string optimizerName = AppConstants.AppArgumentKeys.POLICY_OPTIMIZER.tryRead(processArgs);
if (optimizerName == "")
{
return(null);
}
Type policyOptimizerType = APolicyOptimizer.ChildrenByTypename[optimizerName];
//AppSettings.WriteLogLine("invoking optimizer ...");
APolicyOptimizer opt = (APolicyOptimizer)policyOptimizerType.GetConstructor(new Type[] { }).Invoke(new object[] { });
opt.init(graph, gameParams, processArgs);
opt.process(parallelOptInner);
return(opt);
}
/// <summary>
/// assumes 'processArgs' had a specific optimizer and/or policies, and uses them for the process
/// </summary>
/// <param name="parallelOptInner"></param>
/// <param name="initialProcessArgs">
/// empty values (i.e. keys with empty strings) will be ignored
/// </param>
/// <param name="graph"></param>
/// <param name="optimizeInitialProcessArgs">
/// if false, theoryOutput and optimizerOutput will not be populated
/// and no optimizer will be used to override any of the values in optimizeInitialProcessArgs
/// </param>
/// <returns></returns>
public static ProcessOutput processParams(ParallelOptions parallelOptInner,
Dictionary <string, string> initialProcessArgs,
AGameGraph graph,
bool optimizeInitialProcessArgs = true,
bool calcStdDev = true)
{
ProcessOutput res = new ProcessOutput();
var processArgs = new Dictionary <string, string>(initialProcessArgs);
foreach (var val in initialProcessArgs)
{
if (val.Value == "")
{
processArgs.Remove(val.Key);
}
}
// allocate game process
string gameType = AppConstants.AppArgumentKeys.GAME_MODEL.tryRead(processArgs);
AGameProcess game = Utils.ReflectionUtils.constructEmptyCtorType <AGameProcess>(gameType);
// load game params file. processArgs may override some of the loaded values
IGameParams gameParams = game.constructGameParams();
gameParams.deserialize(AppConstants.AppArgumentKeys.PARAM_FILE_PATH.tryRead(processArgs), processArgs);
APolicyOptimizer preferredOptimizer = null;// null if no optimizer specified
if (optimizeInitialProcessArgs)
{
preferredOptimizer = getOptimizedPolicyInput(parallelOptInner, processArgs, graph, game, gameParams);
}
// if an optimizer was specified, it may have the freedom to choose from several policies, and also provide theoretical bounds *given this freedom*
// note that the optimizer may attempt choosing policies, but can't override values in init file.
string evadersPolicy, pursuersPolicy;
if (preferredOptimizer != null)
{
res.theoryOutput = new Dictionary <string, string>(processArgs);
// FIXME: for creating theoryOutput, if pursuers policy wasn't specificed in input params, we throw an exception.
// this is not necessarily the correct way to do this, since theoretically the optimizer is the one that decides what pursuers policy
// to use
res.theoryOutput.AddRange(getPursuersPolicyTheoreticalOptimizerResult(AppConstants.AppArgumentKeys.PURSUER_POLICY.tryRead(processArgs, ""), graph, gameParams, processArgs));
processArgs.AddRange(preferredOptimizer.optimizationOutput, false);
res.optimizerOutput = new Dictionary <string, string>(processArgs);
res.optimizerOutput.AddRange(preferredOptimizer.optimizationOutput);
}
// evadersPolicy and pursuersPolicy are extracted after the optimizer is used, since the optimizer may choose them
evadersPolicy = AppConstants.AppArgumentKeys.EVADER_POLICY.tryRead(processArgs);
pursuersPolicy = AppConstants.AppArgumentKeys.PURSUER_POLICY.tryRead(processArgs);
// if no optimizer specified, the theoretical results are calculated by the estimator (which assumes specific evaders policy and pursuers policy are given in processArgs)
// similarly to the optimizer, the pursuer's policy can't override previous data in processArgs
if (preferredOptimizer == null && optimizeInitialProcessArgs)
{
res.theoryOutput = new Dictionary <string, string>(processArgs);
var theoreticalOptimizerRes = getPursuersPolicyTheoreticalOptimizerResult(pursuersPolicy, graph, gameParams, processArgs);
res.theoryOutput.AddRange(theoreticalOptimizerRes);
processArgs.AddRange(theoreticalOptimizerRes, false);
}
res.processOutput =
getEstimatedResultsAverage(parallelOptInner, evadersPolicy, pursuersPolicy, gameParams, graph, processArgs, game, calcStdDev);
return(res);
}
