private void goVeryFast_Click(object sender, EventArgs e)
{
qAgent.goveryfast();
interval = 0;
}
private void startLearn(bool delay)
{
if (qAgent == null)
{
var num_inputs = 27; // 9 eyes, each sees 3 numbers (wall, green, red thing proximity)
var num_actions = 5; // 5 possible angles agent can turn
var temporal_window = 4; // amount of temporal memory. 0 = agent lives in-the-moment :)
var network_size = num_inputs * temporal_window + num_actions * temporal_window + num_inputs;
var layer_defs = new List <LayerDefinition>();
// the value function network computes a value of taking any of the possible actions
// given an input state. Here we specify one explicitly the hard way
// but user could also equivalently instead use opt.hidden_layer_sizes = [20,20]
// to just insert simple relu hidden layers.
layer_defs.Add(new LayerDefinition {
type = "input", out_sx = 1, out_sy = 1, out_depth = network_size
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "fc", num_neurons = 96, activation = "relu"
});
layer_defs.Add(new LayerDefinition {
type = "regression", num_neurons = num_actions
});
// options for the Temporal Difference learner that trains the above net
// by backpropping the temporal difference learning rule.
//var opt = new Options { method="sgd", learning_rate=0.01, l2_decay=0.001, momentum=0.9, batch_size=10, l1_decay=0.001 };
//var opt = new Options { method = "adadelta", l2_decay = 0.001, batch_size = 10 };
var opt = new Options {
method = cboLearningMethod.Text, learning_rate = Double.Parse(txtLearningRate.Text), momentum = Double.Parse(txtLearningMomentum.Text), l1_decay = Double.Parse(txtLearningL1Decay.Text), l2_decay = Double.Parse(txtLearningL2Decay.Text), batch_size = Int32.Parse(txtLearningBatch.Text)
};
var tdtrainer_options = new TrainingOptions();
tdtrainer_options.temporal_window = temporal_window;
//tdtrainer_options.experience_size = 30000;
tdtrainer_options.experience_size = experiencesize;
//tdtrainer_options.start_learn_threshold = 1000;
tdtrainer_options.start_learn_threshold = learnthreshold;
tdtrainer_options.gamma = 0.7;
//tdtrainer_options.learning_steps_total = 200000;
tdtrainer_options.learning_steps_total = Int32.Parse(txtLearnTotal.Text);
//tdtrainer_options.learning_steps_burnin = 3000;
tdtrainer_options.learning_steps_burnin = Int32.Parse(txtLearnBurn.Text);
tdtrainer_options.epsilon_min = 0.05;
tdtrainer_options.epsilon_test_time = 0.00;
tdtrainer_options.layer_defs = layer_defs;
tdtrainer_options.options = opt;
// determine when to use shared experience using static
if (chkSharedExperience.Checked && staticExperience)
{
var brain = new DeepQLearnShared(num_inputs, num_actions, tdtrainer_options);
brain.instance = this.instanceNumber;
qAgent = new QAgent(brain, canvas.Width, canvas.Height, Int32.Parse(txtNumberItems.Text), chkRandomItems.Checked, chkObstructItems.Checked, infiniteItems);
}
// determine when to use shared experience using singleton
else if (chkSharedExperience.Checked && !staticExperience)
{
var brain = new DeepQLearnSharedSingleton(num_inputs, num_actions, tdtrainer_options);
brain.instance = this.instanceNumber;
qAgent = new QAgent(brain, canvas.Width, canvas.Height, Int32.Parse(txtNumberItems.Text), chkRandomItems.Checked, chkObstructItems.Checked, infiniteItems);
}
// determine when to use nonshared experience
else
{
var brain = new DeepQLearn(num_inputs, num_actions, tdtrainer_options);
qAgent = new QAgent(brain, canvas.Width, canvas.Height, Int32.Parse(txtNumberItems.Text), chkRandomItems.Checked, chkObstructItems.Checked, infiniteItems);
}
}
else
{
qAgent.startlearn();
}
if (!delay)
{
qAgent.goveryfast();
interval = 0;
}
if (workerThread == null)
{
workerThread = new Thread(new ThreadStart(BackgroundThread));
workerThread.Start();
}
}