Closed maivincent closed 5 years ago
Yeah likely; let me see if there's a fix for this in the new ones.
Good catch.
Fix did induce a new bug:
return np.array(Image.fromarray(observation).resize(shape))
NameError: name 'shape' is not defined
@bhairavmehta95 :
return np.array(Image.fromarray(observation).resize([self.shape[0:1]]))
should be the fix.
Wouldn't it be self.shape
? The splice (means: from index 0 to 1 exclusive, so just 0) above would resize to a single dimension = square of shape[0] x shape[0]
Woops, then it should be self.shape[0:2] sorry. It does not accept the 3 args of self.shape, as resize takes in only 2. I figured that shape being defined as shape=(120, 160, 3))
, only 120 and 160 are needed.
do0o0o0pe thanks!
When running in gym-duckietown conda environment
python -m reinforcement.pytorch.train_reinforcement
I get this error:
I looked a bit and it seems that
imresize
is deprecated in scipy as written here in Backwards incompatible changes: https://github.com/scipy/scipy/releases/tag/v1.3.0rc1Could this be the source of the problem?