|
|
|
|
|
|
|
self.all_log_probs = tf.identity(all_probs, name="action_probs") |
|
|
|
|
|
|
|
single_dim_entropy = 0.5 * tf.reduce_mean( |
|
|
|
tf.log(2 * np.pi * np.e) + tf.square(log_sigma) |
|
|
|
tf.log(2 * np.pi * np.e) + 2 * log_sigma |
|
|
|
) |
|
|
|
# Make entropy the right shape |
|
|
|
self.entropy = tf.ones_like(tf.reshape(mu[:, 0], [-1])) * single_dim_entropy |
|
|
|
|
Ervin Teng