Fix RL tests

src/test/resources/models/reinforcementModel/cartpole/agent/CartPoleDQN.cnnt

@@ -17,7 +17,7 @@ configuration CartPoleDQN {
 
     use_double_dqn : false
 
-    loss : euclidean
+    loss : huber
 
     replay_memory : buffer{
         memory_size : 10000

src/test/resources/models/reinforcementModel/mountaincar/agent/MountaincarCritic.cnna

@@ -8,5 +8,5 @@ implementation Critic(state, action) {
         FullyConnected(units=300)
     ) ->
     Add() ->
-    Relu();
+    Relu()
 }
\ No newline at end of file

src/test/resources/models/reinforcementModel/torcs/agent/dqn/TorcsDQN.cnnt

@@ -23,7 +23,7 @@ configuration TorcsDQN {
 
     use_double_dqn : true
 
-    loss : euclidean
+    loss : huber
 
     replay_memory : buffer{
         memory_size : 1000000

src/test/resources/target_code/gluon/reinforcementModel/cartpole/CNNDataLoader_cartpole_master_dqn.py

@@ -3,8 +3,9 @@ import h5py
 import mxnet as mx
 import logging
 import sys
+from mxnet import nd
 
-class cartpole_master_dqnDataLoader:
+class CNNDataLoader_cartpole_master_dqn:
     _input_names_ = ['state']
     _output_names_ = ['qvalues_label']
 
@@ -14,21 +15,38 @@ class cartpole_master_dqnDataLoader:
     def load_data(self, batch_size):
         train_h5, test_h5 = self.load_h5_files()
 
-        data_mean = train_h5[self._input_names_[0]][:].mean(axis=0)
-        data_std = train_h5[self._input_names_[0]][:].std(axis=0) + 1e-5
+        train_data = {}
+        data_mean = {}
+        data_std = {}
+
+        for input_name in self._input_names_:
+            train_data[input_name] = train_h5[input_name]
+            data_mean[input_name] = nd.array(train_h5[input_name][:].mean(axis=0))
+            data_std[input_name] = nd.array(train_h5[input_name][:].std(axis=0) + 1e-5)
+
+        train_label = {}
+        for output_name in self._output_names_:
+            train_label[output_name] = train_h5[output_name]
+
+        train_iter = mx.io.NDArrayIter(data=train_data,
+                                       label=train_label,
+                                       batch_size=batch_size)
 
-        train_iter = mx.io.NDArrayIter(train_h5[self._input_names_[0]],
-                                       train_h5[self._output_names_[0]],
-                                       batch_size=batch_size,
-                                       data_name=self._input_names_[0],
-                                       label_name=self._output_names_[0])
         test_iter = None
+
         if test_h5 != None:
-            test_iter = mx.io.NDArrayIter(test_h5[self._input_names_[0]],
-                                          test_h5[self._output_names_[0]],
-                                          batch_size=batch_size,
-                                          data_name=self._input_names_[0],
-                                          label_name=self._output_names_[0])
+            test_data = {}
+            for input_name in self._input_names_:
+                test_data[input_name] = test_h5[input_name]
+
+            test_label = {}
+            for output_name in self._output_names_:
+                test_label[output_name] = test_h5[output_name]
+
+            test_iter = mx.io.NDArrayIter(data=test_data,
+                                          label=test_label,
+                                          batch_size=batch_size)
+
         return train_iter, test_iter, data_mean, data_std
 
     def load_h5_files(self):
@@ -36,21 +54,39 @@ class cartpole_master_dqnDataLoader:
         test_h5 = None
         train_path = self._data_dir + "train.h5"
         test_path = self._data_dir + "test.h5"
+
         if os.path.isfile(train_path):
             train_h5 = h5py.File(train_path, 'r')
-            if not (self._input_names_[0] in train_h5 and self._output_names_[0] in train_h5):
-                logging.error("The HDF5 file '" + os.path.abspath(train_path) + "' has to contain the datasets: "
-                              + "'" + self._input_names_[0] + "', '" + self._output_names_[0] + "'")
-                sys.exit(1)
-            test_iter = None
+
+            for input_name in self._input_names_:
+                if not input_name in train_h5:
+                    logging.error("The HDF5 file '" + os.path.abspath(train_path) + "' has to contain the dataset "
+                                  + "'" + input_name + "'")
+                    sys.exit(1)
+
+            for output_name in self._output_names_:
+                if not output_name in train_h5:
+                    logging.error("The HDF5 file '" + os.path.abspath(train_path) + "' has to contain the dataset "
+                                  + "'" + output_name + "'")
+                    sys.exit(1)
+
             if os.path.isfile(test_path):
                 test_h5 = h5py.File(test_path, 'r')
-                if not (self._input_names_[0] in test_h5 and self._output_names_[0] in test_h5):
-                    logging.error("The HDF5 file '" + os.path.abspath(test_path) + "' has to contain the datasets: "
-                                  + "'" + self._input_names_[0] + "', '" + self._output_names_[0] + "'")
-                    sys.exit(1)
+
+                for input_name in self._input_names_:
+                    if not input_name in test_h5:
+                        logging.error("The HDF5 file '" + os.path.abspath(test_path) + "' has to contain the dataset "
+                                      + "'" + input_name + "'")
+                        sys.exit(1)
+
+                for output_name in self._output_names_:
+                    if not output_name in test_h5:
+                        logging.error("The HDF5 file '" + os.path.abspath(test_path) + "' has to contain the dataset "
+                                      + "'" + output_name + "'")
+                        sys.exit(1)
             else:
                 logging.warning("Couldn't load test set. File '" + os.path.abspath(test_path) + "' does not exist.")
+
             return train_h5, test_h5
         else:
             logging.error("Data loading failure. File '" + os.path.abspath(train_path) + "' does not exist.")

src/test/resources/target_code/gluon/reinforcementModel/cartpole/CNNNet_cartpole_master_dqn.py

@@ -101,7 +101,6 @@ class Net_0(gluon.HybridBlock):
             self.fc3_ = gluon.nn.Dense(units=2, use_bias=True)
             # fc3_, output shape: {[2,1,1]}
 
-        self.last_layers['qvalues'] = 'linear'
 
 
     def hybrid_forward(self, F, state):

src/test/resources/target_code/gluon/reinforcementModel/cartpole/CNNTrainer_cartpole_master_dqn.py

@@ -81,7 +81,7 @@ if __name__ == "__main__":
         'qnet':qnet_creator.net,
         'use_fix_target': True,
         'target_update_interval': 200,
-        'loss_function': 'euclidean',
+        'loss': 'huber',
         'optimizer': 'rmsprop',
         'optimizer_params': {
             'learning_rate': 0.001        },

src/test/resources/target_code/gluon/reinforcementModel/cartpole/reinforcement_learning/agent.py

@@ -114,6 +114,8 @@ class Agent(object):
         agent_session_file = os.path.join(session_dir, 'agent.p')
         logger = self._logger
 
+        self._training_stats.save_stats(self._output_directory, episode=self._current_episode)
+
         self._make_pickle_ready(session_dir)
 
         with open(agent_session_file, 'wb') as f:
@@ -177,6 +179,9 @@ class Agent(object):
         return states, actions, rewards, next_states, terminals
 
     def evaluate(self, target=None, sample_games=100, verbose=True):
+        if sample_games <= 0:
+            return 0
+
         target = self._target_score if target is None else target
         if target:
             target_achieved = 0
@@ -268,8 +273,9 @@ class Agent(object):
 
     def _save_net(self, net, filename, filedir=None):
         filedir = self._output_directory if filedir is None else filedir
-        filename = os.path.join(filedir, filename + '.params')
-        net.save_parameters(filename)
+        filename = os.path.join(filedir, filename)
+        net.save_parameters(filename + '.params')
+        net.export(filename, epoch=0)
 
     def save_best_network(self, path, epoch=0):
         self._logger.info(
@@ -367,6 +373,8 @@ class DdpgAgent(Agent):
 
     def _make_pickle_ready(self, session_dir):
         super(DdpgAgent, self)._make_pickle_ready(session_dir)
+        self._save_net(self._actor, 'current_actor')
+
         self._save_net(self._actor, 'actor', session_dir)
         self._actor = None
         self._save_net(self._critic, 'critic', session_dir)
@@ -457,9 +465,9 @@ class DdpgAgent(Agent):
         else:
             self._training_stats = DdpgTrainingStats(episodes)
 
-        # Initialize target Q' and mu'
-        self._actor_target = self._copy_actor()
-        self._critic_target = self._copy_critic()
+            # Initialize target Q' and mu'
+            self._actor_target = self._copy_actor()
+            self._critic_target = self._copy_critic()
 
         # Initialize l2 loss for critic network
         l2_loss = gluon.loss.L2Loss()
@@ -540,7 +548,7 @@ class DdpgAgent(Agent):
                         actor_qvalues = tmp_critic(states, self._actor(states))
                         # For maximizing qvalues we have to multiply with -1
                         # as we use a minimizer
-                        actor_loss = -1 * actor_qvalues
+                        actor_loss = -1 * actor_qvalues.mean()
                     actor_loss.backward()
                     trainer_actor.step(self._minibatch_size)
 
@@ -732,6 +740,7 @@ class DqnAgent(Agent):
 
     def _make_pickle_ready(self, session_dir):
         super(DqnAgent, self)._make_pickle_ready(session_dir)
+        self._save_net(self._qnet, 'current_qnet')
         self._save_net(self._qnet, 'qnet', session_dir)
         self._qnet = None
         self._save_net(self._target_qnet, 'target_net', session_dir)
@@ -897,4 +906,4 @@ class DqnAgent(Agent):
 
     def _save_current_as_best_net(self):
         self._best_net = copy_net(
-            self._qnet, (1,) + self._state_dim, ctx=self._ctx)
+            self._qnet, self._state_dim, ctx=self._ctx)

src/test/resources/target_code/gluon/reinforcementModel/cartpole/reinforcement_learning/strategy.py

@@ -168,5 +168,5 @@ class OrnsteinUhlenbeckStrategy(BaseStrategy):
 
     def select_action(self, values):
         noise = self._evolve_state()
-        action = values + (self.cur_eps * noise)
+        action = (1.0 - self.cur_eps) * values + (self.cur_eps * noise)
         return np.clip(action, self._action_low, self._action_high)

src/test/resources/target_code/gluon/reinforcementModel/cartpole/reinforcement_learning/util.py

@@ -127,13 +127,15 @@ class TrainingStats(object):
         else:
             return self._all_total_rewards[0]
 
-    def save(self, path):
-        np.save(os.path.join(path, 'total_rewards'), self._all_total_rewards)
-        np.save(os.path.join(path, 'eps'), self._all_eps)
-        np.save(os.path.join(path, 'time'), self._all_time)
+    def save(self, path, episode=None):
+        if episode is None:
+            episode = self._max_episodes
+        np.save(os.path.join(path, 'total_rewards'), self._all_total_rewards[:episode])
+        np.save(os.path.join(path, 'eps'), self._all_eps[:episode])
+        np.save(os.path.join(path, 'time'), self._all_time[:episode])
         np.save(
             os.path.join(path, 'mean_reward'),
-            self._all_mean_reward_last_100_episodes)
+            self._all_mean_reward_last_100_episodes[:episode])
 
     def _log_episode(self, episode, start_time, training_steps, eps, reward):
         self.add_eps(episode, eps)
@@ -170,33 +172,43 @@ class DqnTrainingStats(TrainingStats):
         self._logger.info(info)
         return avg_reward
 
-    def save_stats(self, path):
+    def save_stats(self, path, episode=None):
+        if episode is None:
+            episode = self._max_episodes
+
+        all_total_rewards = self._all_total_rewards[:episode]
+        all_avg_loss = self._all_avg_loss[:episode]
+        all_eps = self._all_eps[:episode]
+        all_mean_reward_last_100_episodes = self._all_mean_reward_last_100_episodes[:episode]
+
         fig = plt.figure(figsize=(20, 20))
 
         sub_rewards = fig.add_subplot(221)
         sub_rewards.set_title('Total Rewards per episode')
         sub_rewards.plot(
-            np.arange(self._max_episodes), self._all_total_rewards)
+            np.arange(episode), all_total_rewards)
 
         sub_loss = fig.add_subplot(222)
         sub_loss.set_title('Avg. Loss per episode')
-        sub_loss.plot(np.arange(self._max_episodes), self._all_avg_loss)
+        sub_loss.plot(np.arange(episode), all_avg_loss)
 
         sub_eps = fig.add_subplot(223)
         sub_eps.set_title('Epsilon per episode')
-        sub_eps.plot(np.arange(self._max_episodes), self._all_eps)
+        sub_eps.plot(np.arange(episode), all_eps)
 
         sub_rewards = fig.add_subplot(224)
         sub_rewards.set_title('Avg. mean reward of last 100 episodes')
-        sub_rewards.plot(np.arange(self._max_episodes),
-                         self._all_mean_reward_last_100_episodes)
+        sub_rewards.plot(np.arange(episode),
+                         all_mean_reward_last_100_episodes)
 
-        self.save(path)
+        self.save(path, episode=episode)
         plt.savefig(os.path.join(path, 'stats.pdf'))
 
-    def save(self, path):
-        super(DqnTrainingStats, self).save(path)
-        np.save(os.path.join(path, 'avg_loss'), self._all_avg_loss)
+    def save(self, path, episode=None):
+        if episode is None:
+            episode = self._max_episodes
+        super(DqnTrainingStats, self).save(path, episode=episode)
+        np.save(os.path.join(path, 'avg_loss'), self._all_avg_loss[:episode])
 
 
 class DdpgTrainingStats(TrainingStats):
@@ -233,44 +245,56 @@ class DdpgTrainingStats(TrainingStats):
         self.logger.info(info)
         return avg_reward
 
-    def save(self, path):
-        super(DdpgTrainingStats, self).save(path)
+    def save(self, path, episode=None):
+        if episode is None:
+            episode = self._max_episodes
+        super(DdpgTrainingStats, self).save(path, episode=episode)
         np.save(os.path.join(
-            path, 'avg_critic_loss'), self._all_avg_critic_loss)
-        np.save(os.path.join(path, 'avg_actor_loss'), self._all_avg_actor_loss)
-        np.save(os.path.join(path, 'avg_qvalues'), self._all_avg_qvalues)
+            path, 'avg_critic_loss'), self._all_avg_critic_loss[:episode])
+        np.save(os.path.join(path, 'avg_actor_loss'), self._all_avg_actor_loss[:episode])
+        np.save(os.path.join(path, 'avg_qvalues'), self._all_avg_qvalues[:episode])
+
+    def save_stats(self, path, episode=None):
+        if episode is None:
+            episode = self._max_episodes
+
+        all_total_rewards = self._all_total_rewards[:episode]
+        all_avg_actor_loss = self._all_avg_actor_loss[:episode]
+        all_avg_critic_loss = self._all_avg_critic_loss[:episode]
+        all_avg_qvalues = self._all_avg_qvalues[:episode]
+        all_eps = self._all_eps[:episode]
+        all_mean_reward_last_100_episodes = self._all_mean_reward_last_100_episodes[:episode]
 
-    def save_stats(self, path):
         fig = plt.figure(figsize=(120, 120))
 
         sub_rewards = fig.add_subplot(321)
         sub_rewards.set_title('Total Rewards per episode')
         sub_rewards.plot(
-            np.arange(self._max_episodes), self._all_total_rewards)
+            np.arange(episode), all_total_rewards)
 
         sub_actor_loss = fig.add_subplot(322)
         sub_actor_loss.set_title('Avg. Actor Loss per episode')
         sub_actor_loss.plot(
-            np.arange(self._max_episodes), self._all_avg_actor_loss)
+            np.arange(episode), all_avg_actor_loss)
 
         sub_critic_loss = fig.add_subplot(323)
         sub_critic_loss.set_title('Avg. Critic Loss per episode')
         sub_critic_loss.plot(
-            np.arange(self._max_episodes), self._all_avg_critic_loss)
+            np.arange(episode), all_avg_critic_loss)
 
         sub_qvalues = fig.add_subplot(324)
         sub_qvalues.set_title('Avg. QValues per episode')
         sub_qvalues.plot(
-            np.arange(self._max_episodes), self._all_avg_qvalues)
+            np.arange(episode), all_avg_qvalues)
 
         sub_eps = fig.add_subplot(325)
         sub_eps.set_title('Epsilon per episode')
-        sub_eps.plot(np.arange(self._max_episodes), self._all_eps)
+        sub_eps.plot(np.arange(episode), all_eps)
 
         sub_rewards = fig.add_subplot(326)
         sub_rewards.set_title('Avg. mean reward of last 100 episodes')
-        sub_rewards.plot(np.arange(self._max_episodes),
-                         self._all_mean_reward_last_100_episodes)
+        sub_rewards.plot(np.arange(episode),
+                         all_mean_reward_last_100_episodes)
 
-        self.save(path)
+        self.save(path, episode=episode)
         plt.savefig(os.path.join(path, 'stats.pdf'))

src/test/resources/target_code/gluon/reinforcementModel/mountaincar/CNNCreator_mountaincar_master_actor.py

 import mxnet as mx
 import logging
 import os
-from CNNNet_mountaincar_master_actor import Net
+from CNNNet_mountaincar_master_actor import Net_0
 
 class CNNCreator_mountaincar_master_actor:
     _model_dir_ = "model/mountaincar.agent.MountaincarActor/"
     _model_prefix_ = "model"
-    _input_shapes_ = [(2,)]
 
     def __init__(self):