public CtwContextTreeUndo(MC_AIXI_CTW agent)
{
this.Age = agent.Age;
this.TotalReward = agent.TotalReward;
this.HistorySize = agent.history_size();
this.LastUpdate = agent.LastUpdate;
}
public CtwContextTreeUndo(MC_AIXI_CTW agent)
{
this.Age = agent.Age;
this.TotalReward = agent.TotalReward;
this.HistorySize = agent.history_size();
this.LastUpdate = agent.LastUpdate;
}
public AgentTest()
{
this.Options = new Dictionary<string, string>();
Options["ct-depth"] = "4";
Options["agent-horizon"] = "6";
Options["mc-simulations"] = "200";
this.Env = new CoinFlip(Options);
this.Agent = new MC_AIXI_CTW(Env, Options);
}
private static void Main(string[] args)
{
new ConsoleEnvUI();
return;
var options = new Dictionary <string, string>();
//POSSIBLE OPTIONS:
//ctw-model: possible values ct/ctf/cuda. Which CTW tree implementation should CTW-agent use.
options["ctw-model"] = "ctf";
//exploration: (possible values 0.0-1.0), initial probability of exploration.
// exploration is probability of taking random action
options["exploration"] = "0.99";
//explore-decay (possible values 0.0-1.0). Probability of exploration will decay exponentially.
// It will be multiplied by explore-decay every cycle
options["explore-decay"] = "0.9995";
//ct-depth (possible values int 1+, common values are 8-100) depth of CTW tree.
// That means, how many bits it will look back while deciding
options["ct-depth"] = "96";
//agent-horizon: (values int, 1+) how far into future (in bits) we should try to predict the future.
// From predicted future is computed how good this future is.
options["agent-horizon"] = "4";
//mc-simulations (possible values 1+). Number of monte carlo simulations to do.
// Higher number means better convergence (ie: better data for decision). But also slow speed.
options["mc-simulations"] = "200";
//terminate-age: (possible values int 1+)
// how many cycles to do, before ending. 0=run forever
// cycle = environment gives observation and reward and agent reacts with action
options["terminate-age"] = "10000";
var env = new RockPaperScissorsEnvironment(options);
var agent = new MC_AIXI_CTW(env, options);
var startingTime = DateTime.Now;
InteractionLoop(agent, env, options);
var endingTime = DateTime.Now;
Console.WriteLine("time: {0}", endingTime - startingTime);
Console.ReadLine();
}
private static void Main(string[] args)
{
new ConsoleEnvUI();
return;
var options = new Dictionary<string, string>();
//POSSIBLE OPTIONS:
//ctw-model: possible values ct/ctf/cuda. Which CTW tree implementation should CTW-agent use.
options["ctw-model"] = "ctf";
//exploration: (possible values 0.0-1.0), initial probability of exploration.
// exploration is probability of taking random action
options["exploration"] = "0.99";
//explore-decay (possible values 0.0-1.0). Probability of exploration will decay exponentially.
// It will be multiplied by explore-decay every cycle
options["explore-decay"] = "0.9995";
//ct-depth (possible values int 1+, common values are 8-100) depth of CTW tree.
// That means, how many bits it will look back while deciding
options["ct-depth"] = "96";
//agent-horizon: (values int, 1+) how far into future (in bits) we should try to predict the future.
// From predicted future is computed how good this future is.
options["agent-horizon"] = "4";
//mc-simulations (possible values 1+). Number of monte carlo simulations to do.
// Higher number means better convergence (ie: better data for decision). But also slow speed.
options["mc-simulations"] = "200";
//terminate-age: (possible values int 1+)
// how many cycles to do, before ending. 0=run forever
// cycle = environment gives observation and reward and agent reacts with action
options["terminate-age"] = "10000";
var env = new RockPaperScissorsEnvironment(options);
var agent = new MC_AIXI_CTW(env, options);
var startingTime = DateTime.Now;
InteractionLoop(agent, env, options);
var endingTime = DateTime.Now;
Console.WriteLine("time: {0}", endingTime - startingTime);
Console.ReadLine();
}
