Fix for divide-by-zero error with Discrete Actions (#1520)

* Enable buffer padding to be set other than 0

Allows buffer padding in AgentBufferField to be set to a custom value. In particular, 0-padding for `action_masks` causes a divide-by-zero error, and should be padded with 1’s instead.

This is done as a parameter passed to the `append` method, so that the pad value can be set right after the instantiation of an AgentBufferField.

ml-agents/mlagents/trainers/buffer.py

             AgentBufferField with the append method.
             """
 
+            def __init__(self):
+                self.padding_value = 0
+                super(Buffer.AgentBuffer.AgentBufferField, self).__init__()
+
+            def append(self, element, padding_value=0):
+                """
+                Adds an element to this list. Also lets you change the padding 
+                type, so that it can be set on append (e.g. action_masks should
+                be padded with 1.) 
+                :param element: The element to append to the list.
+                :param padding_value: The value used to pad when get_batch is called.
+                """
+                super(Buffer.AgentBuffer.AgentBufferField, self).append(element)
+                self.padding_value = padding_value
+
-                Ads a list of np.arrays to the end of the list of np.arrays.
+                Adds a list of np.arrays to the end of the list of np.arrays.
                 :param data: The np.array list to append.
                 """
                 self += list(np.array(data))
                             raise BufferException("The batch size and training length requested for get_batch where"
                                                   " too large given the current number of data points.")
                         tmp_list = []
-                        padding = np.array(self[-1]) * 0
+                        padding = np.array(self[-1]) * self.padding_value
                         # The padding is made with zeros and its shape is given by the shape of the last element
                         for end in range(len(self), len(self) % training_length, -training_length)[:batch_size]:
                             tmp_list += [np.array(self[end - training_length:end])]

ml-agents/mlagents/trainers/ppo/trainer.py

                             epsilons[idx])
                     else:
                         self.training_buffer[agent_id]['action_mask'].append(
-                            stored_info.action_masks[idx])
+                            stored_info.action_masks[idx], padding_value=1)
                     a_dist = stored_take_action_outputs['log_probs']
                     value = stored_take_action_outputs['value']
                     self.training_buffer[agent_id]['actions'].append(actions[idx])