ValueError: Expected positive integer total_steps, but got -1

[morestart]

    def configure_optimizers(self):
        optimizer = torch.optim.SGD(self.parameters(), lr=self.lr)
        print(self.trainer.max_steps)
        lr_scheduler = {
            'scheduler': torch.optim.lr_scheduler.OneCycleLR(optimizer,
                                                             max_lr=self.lr,
                                                             total_steps=self.trainer.max_steps,
                                                             anneal_strategy='linear',
                                                             cycle_momentum=False,
                                                             pct_start=0.1),
            'interval': 'step',
            'frequency': 1
        }
        return {'optimizer': optimizer, 'lr_scheduler': lr_scheduler, "monitor": 'val_acc'}

raise ValueError("Expected positive integer total_steps, but got {}".format(total_steps))
ValueError: Expected positive integer total_steps, but got -1

[morestart]

ok i know the answer,

[eyildiz-ugoe]

Well, would you care to let us know?

[rohitgr7]

well,

self.trainer.max_steps

max_steps doesn't tell total optimization steps if you don't set it.

There's feature PR that can help you with this once it get's merged: #11599

[morestart]

great work, this is my way to compute max_steps:

max_steps = epochs * int(np.ceil(len(self._build_dataloader().train_dataloader()) / 1))