Warning in torch.lr_scheduler

[CrafterKolyan]

🐛 Bug Report

On catalyst training with scheduler a warning shows.

OS: Windows
Python 3.7.8
catalyst==20.10.1
torch==1.6.0

How To Reproduce

import torch
from torch import nn
import torch.nn.functional as F
from torch.utils.data import TensorDataset, DataLoader
from catalyst import dl

X = torch.randn(1000, 10)
y = torch.rand(X.shape[0])
model = nn.Linear(X.shape[1], 1)
optimizer = torch.optim.Adam(model.parameters())
scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.9)

dataset = TensorDataset(X, y)

loaders = {
    'train': DataLoader(dataset, batch_size=32, shuffle=True),
    'valid': DataLoader(dataset, batch_size=32)
}

class CustomRunner(dl.SupervisedRunner):
    def _handle_batch(self, batch):
        # model train/valid step
        y_pred = self.model(batch['features']).view(-1)

        loss = F.mse_loss(y_pred, batch['targets'])
        self.batch_metrics.update(
            {"loss": loss}
        )

runner = CustomRunner()

runner.train(
    model=model,
    loaders=loaders,
    optimizer=optimizer,
    scheduler=scheduler,
    num_epochs=20
)

Screenshots

Expected behavior

No warnings

PS

• I know, that I could join Catalyst slack (#__questions channel) for issue discussion.

[and-kul]

#980 has chances to fix this problem

[CrafterKolyan]

Should be fixed now