37 lines
2.3 KiB
Python
37 lines
2.3 KiB
Python
import tensorflow as tf
|
|
|
|
|
|
def loss(nextState, actions):
|
|
# return tf.reduce_sum(tf.square(nextState[:, 2:, 0] * (0.5 * (nextState[:, 0] + 0.25 * nextState[:, 1] - actions))), axis=1)
|
|
return tf.reduce_mean(tf.square(nextState[:, 0] + 0.25 * nextState[:, 1] - tf.reduce_sum(
|
|
nextState[:, 2:6, 0] * (actions[:, :4] + actions[:, 4:]), axis=1))) + tf.reduce_mean(
|
|
tf.reduce_sum(tf.square(nextState[:, 6:, 0] - actions[:, :4]), axis=1), axis=0)
|
|
|
|
|
|
def loss2(nextState, actions):
|
|
# return tf.reduce_sum(tf.square(nextState[:, 2:, 0] * (0.5 * (nextState[:, 0] + 0.25 * nextState[:, 1] - actions))), axis=1)
|
|
|
|
# return 0.1 * tf.reduce_mean(tf.square(0.75 * nextState[:, 1] - tf.reduce_sum(nextState[:, 2:6, 0] * (actions[:, 4:] + actions[:, :4]),axis=1))) + 0.9 * tf.reduce_mean(tf.reduce_sum(tf.square(nextState[:, 6:, 0] - actions[:, :4]), axis=1), axis=0)
|
|
|
|
# return 0.0 * tf.reduce_mean(tf.square(0.75 * nextState[:, 1] - tf.reduce_sum(nextState[:, 2:6, 0] * (actions[:, :4]),axis=1))) + 1.0 * tf.reduce_mean(tf.reduce_sum(tf.square(nextState[:, 6:, 0] - actions[:, :4]), axis=1), axis=0)
|
|
|
|
return tf.reduce_mean(
|
|
tf.reduce_max(nextState[:, 2:6, 0] * tf.square((nextState[:, 6:, 0] - (actions[:, :4] + actions[:, 4:]))),
|
|
axis=1), axis=0)
|
|
|
|
# action = nextState[:, 3] * 1 + nextState[:, 4] * 2 + nextState[:, 5] * 3
|
|
# action = tf.cast(action, tf.int32)
|
|
# action = tf.reshape(action, (-1,))
|
|
#
|
|
# # test = actions[:, action[:]]
|
|
#
|
|
# test1 = tf.slice(actions[:, :4], action, (-1, 1))
|
|
# test2 = tf.slice(actions[:, 4:], action, (-1, 1))
|
|
#
|
|
# return 1.0 * tf.reduce_mean(tf.reduce_sum(tf.square((0.1 * nextState[:, 1] + nextState[:, 6:, 0]) - (test1 + test2)), axis=1)) + 0.0 * tf.reduce_mean(tf.reduce_sum(tf.square(nextState[:, 6:, 0] - actions[:, :4]), axis=1), axis=0)
|
|
# return 1.0 * tf.reduce_mean(tf.reduce_sum(tf.square((0.1 * nextState[:, 1] + nextState[:, 6:, 0]) - (actions[:, :4] + actions[:, 4:])), axis=1)) + 0.0 * tf.reduce_mean(tf.reduce_sum(tf.square(nextState[:, 6:, 0] - actions[:, :4]), axis=1), axis=0)
|
|
|
|
|
|
def loss3(target, pred):
|
|
return tf.reduce_mean(0.5 * tf.square(0.1 * target[:, 0, 0] + target[:, 1, 0] - (pred[:, 0] + pred[:, 1]))
|
|
+ 0.5 * tf.square(target[:, 1, 0] - pred[:, 0]), axis=0)
|