search

jjkke88 / RL_toolbox

reinfore learning tool box, contains trpo, a3c algorithm for continous action space
MIT License
43 stars 8 forks source link

Unable to train model #1

Open abhinavrai44 opened 7 years ago

abhinavrai44 commented 7 years ago

On executing trpo_continous.py, I get the following error:

[2017-07-01 23:52:58,375] Making new env: CartPole-v0 [TL] InputLayer continous_shared/continous_input_layer: (?, 3) [TL] DenseLayer continous_shared/continous_fc1: 64 relu [TL] DenseLayer continous_shared/continous_fc2: 64 relu [TL] DenseLayer continous_shared/continous_fc3: 1 relu

** Iteration 0 **** Traceback (most recent call last): File "experiment/trpo_continous.py", line 62, in agent.learn() File "/home/abhinav/Desktop/major/parallel-trpo/RL_toolbox/RLToolbox/agent/TRPO_agent.py", line 80, in learn stats , theta , thprev = self.train_mini_batch(linear_search=False) File "/home/abhinav/Desktop/major/parallel-trpo/RL_toolbox/RLToolbox/algorithm/TRPO.py", line 62, in train_mini_batch self.get_samples(self.pms.paths_number) File "/home/abhinav/Desktop/major/parallel-trpo/RL_toolbox/RLToolbox/algorithm/TRPO.py", line 29, in get_samples self.storage.get_single_path() File "/home/abhinav/Desktop/major/parallel-trpo/RL_toolbox/RLToolbox/storage/storage_continous.py", line 36, in get_single_path a, agent_info = self.agent.get_action(o) File "/home/abhinav/Desktop/major/parallel-trpo/RL_toolbox/RLToolbox/algorithm/TRPO.py", line 43, in get_action {self.net.obs: obs}) File "/home/abhinav/anaconda2/envs/osim/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 710, in run run_metadata_ptr) File "/home/abhinav/anaconda2/envs/osim/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 887, in _run % (np_val.shape, subfeed_t.name, str(subfeed_t.get_shape()))) ValueError: Cannot feed value of shape (1, 4) for Tensor u'continous_shared/continous_obs:0', which has shape '(?, 3)'

jjkke88 commented 7 years ago

it seems that the state shape you use is 4 but i know the carpole's state is 3 dimensions, maybe you use a high version of gym or wrong environment