pytorch_coma
pytorch_coma copied to clipboard
Serious problem with data normalize
data.py, Line91-99 The inherited attribute is 'pre_transform', while during __init__ another 'pre_tranform'(tranform v.s. tranSform) is initialized.
Here only the mean and std of 'pre_tranform'(no S) is initialized, but later used is 'pre_transform'(with S)
In this way, a serious normalize problem happens here.
if self.pre_transform is not None:
if hasattr(self.pre_transform, 'mean') and hasattr(self.pre_transform, 'std'):
if self.pre_tranform.mean is None:
self.pre_tranform.mean = mean_train # no S
if self.pre_transform.std is None:
self.pre_tranform.std = std_train # no S
train_data = [self.pre_transform(td) for td in train_data] # with S
val_data = [self.pre_transform(td) for td in val_data]
test_data = [self.pre_transform(td) for td in test_data]
https://github.com/pixelite1201/pytorch_coma/blob/8446eec576c80b2d9338a04ec5a9b465ae2f1137/data.py?_pjax=%23js-repo-pjax-container%2C%20div%5Bitemtype%3D%22http%3A%2F%2Fschema.org%2FSoftwareSourceCode%22%5D%20main%2C%20%5Bdata-pjax-container%5D#L91-L99