fixed continuous action explorations

krfricke · krfricke · commit a0c603355bb7 · 2018-02-18T22:40:33.000+01:00
diff --git a/tensorforce/core/explorations/__init__.py b/tensorforce/core/explorations/__init__.py
@@ -16,19 +16,17 @@
 
 from tensorforce.core.explorations.exploration import Exploration
 from tensorforce.core.explorations.constant import Constant
-from tensorforce.core.explorations.linear_decay import LinearDecay
 from tensorforce.core.explorations.epsilon_anneal import EpsilonAnneal
 from tensorforce.core.explorations.epsilon_decay import EpsilonDecay
 from tensorforce.core.explorations.ornstein_uhlenbeck_process import OrnsteinUhlenbeckProcess
 
 
 explorations = dict(
     constant=Constant,
-    linear_decay=LinearDecay,
     epsilon_anneal=EpsilonAnneal,
     epsilon_decay=EpsilonDecay,
     ornstein_uhlenbeck=OrnsteinUhlenbeckProcess
 )
 
 
-__all__ = ['Exploration', 'Constant', 'LinearDecay', 'EpsilonDecay', 'OrnsteinUhlenbeckProcess', 'explorations']
+__all__ = ['Exploration', 'Constant', 'EpsilonAnneal', 'EpsilonDecay', 'OrnsteinUhlenbeckProcess', 'explorations']
diff --git a/tensorforce/core/explorations/linear_decay.py b/tensorforce/core/explorations/linear_decay.py
diff --git a/tensorforce/core/explorations/ornstein_uhlenbeck_process.py b/tensorforce/core/explorations/ornstein_uhlenbeck_process.py
@@ -44,6 +44,10 @@ def __init__(
 
     def tf_explore(self, episode, timestep, action_shape):
         normal_sample = tf.random_normal(shape=action_shape.shape, mean=0.0, stddev=1.0)
-        state = tf.get_variable(name='ornstein_uhlenbeck', dtype=util.tf_dtype('float'), shape=action_shape.shape,
-                                initializer=tf.constant_initializer(self.mu))
+        state = tf.get_variable(
+            name='ornstein_uhlenbeck',
+            dtype=util.tf_dtype('float'),
+            shape=action_shape.shape,
+            initializer=tf.constant_initializer(self.mu)
+        )
         return tf.assign_add(ref=state, value=(self.theta * (self.mu - state) + self.sigma * normal_sample))
diff --git a/tensorforce/models/model.py b/tensorforce/models/model.py
@@ -847,7 +847,7 @@ def tf_action_exploration(self, action, exploration, action_spec):
 
         elif action_spec['type'] == 'float':
             for _ in range(util.rank(action) - 1):
-                exploration_value = tf.expand_dims(input=exploration_value, axis=1)
+                exploration_value = tf.expand_dims(input=exploration_value, axis=-1)
             action += exploration_value
             if 'min_value' in action_spec:
                 action = tf.clip_by_value(
