RicardoDominguez / PyCREPS

Contextual Relative Entropy Policy Search for Reinforcement Learning in Python
15 stars 1 forks source link

Error cartPole with use_torch = True #17

Closed RicardoDominguez closed 6 years ago

RicardoDominguez commented 6 years ago

Running cartPole\cartPole_learn.py with use_torch = True

Traceback (most recent call last): File "pj\PyCREPS\cartPole\cartPole_learn.py", line 73, in R, W, F = predictReward(env, M, hpol, pol) File "C:\Users\Marijose\pj\PyCREPS\cartPole\scenario.py", line 53, in predictReward w = hipol.sample(s.reshape(1, -1)) # Sample lower-policy weights File "C:\Users\Marijose\pj\PyCREPS\CREPS_torch.py", line 182, in sample W[sample, :] = self.mvnrnd.sample() File "C:\Users\Marijose\Miniconda3\envs\th\lib\site-packages\torch\distributions\distribution.py", line 97, in sample return self.rsample(sample_shape) File "C:\Users\Marijose\Miniconda3\envs\th\lib\site-packages\torch\distributions\multivariatenormal.py", line 175, in rsample eps = self.loc.new(*shape).normal() AttributeError: 'numpy.ndarray' object has no attribute 'new'