How to find the current epoch number inside the training loop?

❓ Questions and Help

How to find the current epoch number inside the training loop?

I am training a Conv autoencoder on MNIST images and want to save the reconstructed images after every 10 epochs.

Code

class AutoEncoder(pl.LightningModule):
    def __init__(self):
        pass
    def forward(self, x):
        pass
    def training_step(self, batch, batch_idx):
         images = batch
         out  = self.forward(images)
         loss = self.loss(out, images)
         if epoch % 10:
            save(images)

I am looking for something like this where the output images are saved every 10 epoch.

What have you tried?

I haven’t tried anything since I am unable to find any documentation on current epoch no.

What’s your environment?

I am using pytorch 1.4 and Lightning version 0.7

• OS: [e.g. iOS, Linux, Win]
• Packaging [e.g. pip, conda]
• Version [e.g. 0.5.2.1]

[philip30]

self.current_epoch would do. (self is a lightning module). Though it starts with 0. You can find the definition of the module here.

https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/core/lightning.py

Though I’d recommend you to use IDE e.g. pycharm so you can go to the definition of the class quickly.

[pme0]

In case the iteration/step is required, there is also self.global_step