def train(self, batch):
batch_size = batch['batch_size']
target = self.get_target(batch)
actions = np.zeros([batch_size, self.num_actions])
actions[:, list(batch['actions'])] = 1
_, error_val = self.session.run(
[self.train_step, self.error],
feed_dict=get_feed_dict([(self.target_pl_list, target), (self.actions_pl_list, actions),
(self.input_pl_list, batch['states_1'])], len(self.gpus))
)
return error_val
def evaluate_target(self, batch):
return self.session.run(self.target_out, feed_dict=get_feed_dict([(self.input_pl_list, batch)], len(self.gpus)))
