Migrating from PyTorch

Joan · February 10, 2020, 2:05pm

Hi,

I am trying to use the Migrating notebook from fastai2 to try to use some PyTorch code in fastai2.

It seems that everything work until I tried to fit_one_cycle when I got an error about pbar . Disable it does not seem to solve the issue. Here is the code with the error.

I did a fresh install of fastai2, openslide-python (1.1.1) and fastprogress (0.2.2)

Any ideas how to solve the problem? Thanks

P.S.: This question was first here but I opened a new issue bc is kind of unrelated

sgugger · February 10, 2020, 6:19pm

It looks like the DataLoader could not pull the len of your dataset. Can you try len(ds) (where ds is what you give the DataLoader)?

Joan · February 11, 2020, 11:27am

Thanks @sgugger for the answer. It seems that the custom dataset build has an argument (mode) that, if not specified, does not generate the ds. However, things seems a little bit more complex than default migrating. I made a summary of what I think it does:

BACKGROUND: This pipeline uses openslide library to tile a very big image (WSI - a tissue histology image) and make a prediction in a weakly supervised way. The authors mean with this that they only have an overall label rather than a per tile label. The workflow goes more or less like:

Perform predictions directly on all the tiles generated and got probabilities
Reorder the tiles with the greatest probability
Keep only the tiles with the greatest probability and generate a subset of images with the overall label (converting the problem to a fully supervised fashion)
Perfrom training on this subset and update weights and optimizer

So, if I understood the code correctly, this pipeline perform a first prediciton step that is not suitable for fastai2 Dataloaders to work out of the box. Am I right? Do you think is there any possibility to implement this kind of pipeline in fastai2? Maybe is just enough to give the dataloader a first prediction and then the subsequent steps in the fastai2 pipeline will work?
Thanks

chr · June 13, 2021, 11:33am

Hi,
I am trying to use a PyTorch model with fastai2 in Google Colab using the Migration tool from

using:
from migrating_pytorch import *

raises the error: ModuleNotFoundError: No module named ‘migrating_pytorch’

Which also happened when running the Tutorial on Colab

Thank you for any help

florianl · June 13, 2021, 5:09pm

I guess the migrating_pytorch.py is missing. you can find them here:

chr · June 13, 2021, 5:24pm

Thank you,
beginner’s question: how do I load / download / integrate such a code from GitHub to Google Colab?
I couldn’t work it out myself when I found this page before posting my question.
Thanks!

florianl · June 13, 2021, 5:52pm

Just add the the following line at the top of the notebook in Colab:

!wget https://raw.githubusercontent.com/fastai/fastai/master/nbs/examples/migrating_pytorch.py

you can run os commands in Jupiter notebooks with !<oscommand>

you should read the code in the .py file … the actual code is in there instead of the notebook imho.

github.com

fastai/fastai/blob/master/nbs/examples/migrating_pytorch.py

import torch
from torch import nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
from torchvision import datasets, transforms

class Flatten(nn.Module):
    def forward(self, x): return x.view(x.size(0), -1)

class Net(nn.Sequential):
    def __init__(self):
        super().__init__(
            nn.Conv2d(1, 32, 3, 1), nn.ReLU(),
            nn.Conv2d(32, 64, 3, 1), nn.MaxPool2d(2), nn.Dropout2d(0.25),
            Flatten(), nn.Linear(9216, 128), nn.ReLU(), nn.Dropout2d(0.5),
            nn.Linear(128, 10), nn.LogSoftmax(dim=1) )

def train(model, device, train_loader, optimizer, epoch):
    model.train()
    for batch_idx, (data, target) in enumerate(train_loader):

This file has been truncated. show original

chr · June 14, 2021, 9:29pm

Thank you
The migration part worked, but the learning / fit one cycle with the PyTorch model not yet.
Will have to figure that out.
Best regards