AssertionError in visualizing trained models.

[gcruiser]

Hi! Thanks for the excellent work!
After I used train_all_copo_dist.py to train the model and try to visualize it, I found that the dimensions of OBS and Weights did not match, even the same problem in the <new_best_checkpoints>. This problem occurred in every environment I trained, and the number of rows in OBS is always 1 less than the number of columns in Weights.
But I can visualize the <new_best_checkpoints> by the new_vis.py, which is wired.

Traceback (most recent call last):
  File "/home/CoPO_tf/CoPO/copo_code/copo/vis_from_checkpoint.py", line 91, in <module>
    action = policy_function(o, d)
  File "/home/CoPO_tf/CoPO/copo_code/copo/eval/get_policy_function.py", line 170, in __call__
    actions = self.policy(obs_to_be_eval)
  File "/home/CoPO_tf/CoPO/copo_code/copo/vis_from_checkpoint.py", line 72, in policy
    ret = policy_class(weights, obs, policy_name=policy_name, layer_name_suffix="_1", deterministic=deterministic)
  File "/home/CoPO_tf/CoPO/copo_code/copo/eval/get_policy_function.py", line 61, in _compute_actions_for_tf_policy
    assert obs.shape[1] == weights[s].shape[0], (obs.shape, weights[s].shape)
AssertionError: ((20, 96), (97, 256))

[pengzhenghao]

The problem must come from the missing LCF dimension. In CoPO we add one extra dimension to the obs to include the social orientation.Best,ZhenghaoOn Mar 8, 2024, at 04:06, gcruiser ***@***.***> wrote: Hi! Thanks for the excellent work! After I used train_all_copo_dist.py to train the model and try to visualize it, I found that the dimensions of OBS and Weights did not match, even the same problem in the <new_best_checkpoints>. This problem occurred in every environment I trained, and the number of rows in OBS is always 1 less than the number of columns in Weights. But I can visualize the <new_best_checkpoints> by the new_vis.py, which is wired. Traceback (most recent call last): File "/home/CoPO_tf/CoPO/copo_code/copo/vis_from_checkpoint.py", line 91, in <module> action = policy_function(o, d) File "/home/CoPO_tf/CoPO/copo_code/copo/eval/get_policy_function.py", line 170, in __call__ actions = self.policy(obs_to_be_eval) File "/home/CoPO_tf/CoPO/copo_code/copo/vis_from_checkpoint.py", line 72, in policy ret = policy_class(weights, obs, policy_name=policy_name, layer_name_suffix="_1", deterministic=deterministic) File "/home/CoPO_tf/CoPO/copo_code/copo/eval/get_policy_function.py", line 61, in _compute_actions_for_tf_policy assert obs.shape[1] == weights[s].shape[0], (obs.shape, weights[s].shape) AssertionError: ((20, 96), (97, 256)) —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>