Simplify TD3 Reinforcement Learning Agent Creation

src/mlpack/methods/reinforcement_learning/reinforcement_learning.hpp

@@ -12,6 +12,8 @@
 #ifndef MLPACK_METHODS_REINFORCEMENT_LEARNING_REINFORCEMENT_LEARNING_HPP
 #define MLPACK_METHODS_REINFORCEMENT_LEARNING_REINFORCEMENT_LEARNING_HPP
 
+#include "td3_default_values.hpp"
+
 #include "environment/environment.hpp"
 #include "policy/policy.hpp"
 #include "q_networks/q_networks.hpp"

src/mlpack/methods/reinforcement_learning/td3.hpp

@@ -19,6 +19,7 @@
 
 #include "replay/replay.hpp"
 #include "training_config.hpp"
+#include "td3_default_values.hpp"
 
 namespace mlpack {
 
@@ -47,9 +48,9 @@ namespace mlpack {
  */
 template <
   typename EnvironmentType,
-  typename QNetworkType,
-  typename PolicyNetworkType,
-  typename UpdaterType,
+  typename QNetworkType=decltype(DefaultValues::qNetwork()),
+  typename PolicyNetworkType=decltype(DefaultValues::policyNetwork()),
+  typename UpdaterType=decltype(DefaultValues::default_updater),
   typename ReplayType = RandomReplay<EnvironmentType>
 >
 class TD3

src/mlpack/methods/reinforcement_learning/td3_default_values.hpp

@@ -0,0 +1,69 @@
+/**
+ * @file methods/reinforcement_learning/td3_default_values.hpp
+ * @author Advaith P
+ *
+ * This file has default values for the TD3 class,which implements the 
+ * Twin Delayed Deep Deterministic Policy Gradient algorithm.
+ *
+ * mlpack is free software; you may redistribute it and/or modify it under the
+ * terms of the 3-clause BSD license.  You should have received a copy of the
+ * 3-clause BSD license along with mlpack.  If not, see
+ * http://www.opensource.org/licenses/BSD-3-Clause for more information.
+ */
+#ifndef DEFAULT_VALUES_RL_TD3
+#define DEFAULT_VALUES_RL_TD3
+
+
+
+#include <mlpack/core.hpp>
+#include <ensmallen.hpp>
+#include <mlpack/methods/ann/ann.hpp>
+#include "training_config.hpp"
+
+using namespace mlpack;
+using namespace ens;
+
+
+namespace DefaultValues 
+{
+
+inline TrainingConfig TrainingConfigWithDefaults() {
+    TrainingConfig config;
+    config.StepSize() = 0.01;
+    config.TargetNetworkSyncInterval() = 1;
+    config.UpdateInterval() = 3;
+    config.Rho() = 0.001;
+    return config;
+}
+
+inline FFN<EmptyLoss, GaussianInitialization> qNetwork() {
+    // Construct qNetwork instance
+    FFN<EmptyLoss, GaussianInitialization> qNetworkInstance(EmptyLoss(), GaussianInitialization(0, 0.1));
+    // Add layers
+    qNetworkInstance.Add(new Linear(128));
+    qNetworkInstance.Add(new ReLU());
+    qNetworkInstance.Add(new Linear(1));
+    return qNetworkInstance;
+}
+
+inline FFN<EmptyLoss, GaussianInitialization> policyNetwork() {
+    // Construct policyNetwork instance
+    FFN<EmptyLoss, GaussianInitialization> policyNetworkInstance(EmptyLoss(), GaussianInitialization(0, 0.1));
+    // Add layers
+    policyNetworkInstance.Add(new Linear(128));
+    policyNetworkInstance.Add(new ReLU());
+    policyNetworkInstance.Add(new Linear(1));
+    policyNetworkInstance.Add(new TanH());
+    return policyNetworkInstance;
+}
+
+
+AdamUpdate default_updater;
+
+
+
+
+}
+#endif
+
+