|
|
|
|
|
|
|
|
|
|
|
self.action_holder = tf.placeholder( |
|
|
|
shape=[None, len(policy_branches)], dtype=tf.int32, name="action_holder") |
|
|
|
self.selected_actions = tf.concat([ |
|
|
|
self.action_oh = tf.concat([ |
|
|
|
self.selected_actions = tf.stop_gradient(self.action_oh) |
|
|
|
|
|
|
|
self.all_old_log_probs = tf.placeholder( |
|
|
|
shape=[None, sum(self.act_size)], dtype=tf.float32, name='old_probabilities') |
|
|
|
|
|
|
|
|
|
|
|
self.log_probs = tf.reduce_sum((tf.stack([ |
|
|
|
-tf.nn.softmax_cross_entropy_with_logits_v2( |
|
|
|
labels=self.selected_actions[:, action_idx[i]:action_idx[i + 1]], |
|
|
|
labels=self.action_oh[:, action_idx[i]:action_idx[i + 1]], |
|
|
|
labels=self.selected_actions[:, action_idx[i]:action_idx[i + 1]], |
|
|
|
labels=self.action_oh[:, action_idx[i]:action_idx[i + 1]], |
|
|
|
logits=old_normalized_logits[:, action_idx[i]:action_idx[i + 1]] |
|
|
|
) |
|
|
|
for i in range(len(self.act_size))], axis=1)), axis=1, keepdims=True) |
vincentpierre