Reshape(): argument 'input' (position 1) must be Tensor, not numpy.ndarray

reinforcement-learning
1 
@override(ModelV2)
def forward(self, input_dict, states, seq_lens):
    obs_a = input_dict["prev_n_obs_a"]
    obs_b = input_dict["prev_n_obs_b"]

    obs_a = torch.reshape(obs_a, [-1, obs_space_a.shape[0] * self.num_frames])
    obs_b = torch.reshape(obs_b, [-1, obs_space_b.shape[0] * self.num_frames])

ERROR

in __call__
    res = self.forward(restored, state or [], seq_lens)
  File "/home/niss/proj/lotlatp/pbmain.py", line 116, in forward
    obs_a = torch.reshape(obs_a, [-1, obs_space_a.shape[0] * self.num_frames])
TypeError: reshape(): argument 'input' (position 1) must be Tensor, not numpy.ndarray
2

Did you try to pass a tensor instead of a numpy.array as described in the error message?
You can create tensors from numpy.arrays via tensor = torch.from_numpy(array).