rohit_gr
(Rohit Gupta)
1
train_interp = ClassificationInterpretation.from_learner(learn, ds_type=DatasetType.Train)
train_interp.plot_top_losses(9)
plot_top_losses is still hardcoded with valid_ds
Replacing with
self.data.dl(self.ds_type).dataset[idx]
might work for both.
rohit_gr
(Rohit Gupta)
3
Ok.
Just one more question @sgugger. resize doesn’t work if tfms=None?
data = (src.transform(tfms=None, size=150, resize_method=ResizeMethod.SQUISH)
.databunch(bs=32))
It gives error with learn.lr_find()
:
RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0. Got 150 and 113 in dimension 2
I want to try on raw_images with no other transformations except resizing.
Do I need to create a get_transforms() with proper values in order to avoid transformations or is there any other way to do just resizing?
sgugger
4
It was a bug fixed yesterday actually, so you won’t have the problem in master (or v1.0.43 when it’s released).