Is training steps relevant with batch size?

[fortunechen]

Hi,

Thx for your open-source code !

I find that training steps per epoch is not relevant with batch size. In line 343

steps_per_epoch = num_train_examples // (jax.local_device_count() * config.d_step_per_g_step)

maybe it shoud be

steps_per_epoch = num_train_examples // (jax.local_device_count() * config.d_step_per_g_step * config.batch_size)

[kohjingyu]

Thanks, you are correct. However, we don't use steps_per_epoch in the training loop (we only use num_train_steps rather than counting epochs), so this does not affect training.

[jayati-naik]

Hi,

But num_train_steps are calculated based on steps_per_epoch.

xmcgan_image_generation/xmcgan/train_utils.py

Line 345 in 22a7ef2

num_train_steps = steps_per_epoch * config.num_epochs

In my case with 1000 training examples and 200 validation example, if I go for
batch_size = 8 , d_step_per_g_step = 2
num_train_steps = 2500 and it fails around step 1600 with OutOfRangeError at below step.

xmcgan_image_generation/xmcgan/train_utils.py

Line 421 in 22a7ef2

batch = jax.tree_map(np.asarray, next(train_iter))

Am I missing something here?