I get this error (ValueError: The parameter loc has invalid values) that aborts training of my agent. The environment is openai's pendulum

Question

The environment is Pendulum from OpenAI Gym. Pytorch 1.9.1+cu102 is the package used for training the model.

This error occurs around the code :

out = Lorian(torch.tensor(observation))
distro = torch.distributions.Normal(out[0],torch.absolute(out[1]))

Error :

Traceback (most recent call last):
  File "D:\pythons\pendulum\Main1.py", line 90, in <module>
    distro = torch.distributions.Normal(out[0],torch.absolute(out[1]))
  File "C:\Users\Mukundan1\AppData\Local\Programs\Python\Python39\lib\site-packages\torch\distributions\normal.py", line 50, in __init__
    super(Normal, self).__init__(batch_shape, validate_args=validate_args)
  File "C:\Users\Mukundan1\AppData\Local\Programs\Python\Python39\lib\site-packages\torch\distributions\distribution.py", line 53, in __init__
    raise ValueError("The parameter {} has invalid values".format(param))
ValueError: The parameter loc has invalid values

the object 'distro' is sampled to get the agent's decision given the current observation. Lorian is the name of this model/neural network. The output of the neural network can be either positive and negative but the standard deviation from mean can only be a positive since torch.absolute(out[1]) is used.

Before I started using the absolute value of out[1] the error was mostly the same except instead of 'loc' it said 'scale'.

Where is the mistake in my input of torch.distributions.Normal()?

Answer 1

you can "print" the output of the neural network.May be your output is 'nan' that case the error.