What is the difference between steps and epochs in TensorFlow?

Question

I have noticed that in most models step parameter is used to indicate the no. of steps to run over data but I have also noticed that in practical usage, we mainly use the fit function N epochs.

What is the distinction between running one thousand steps with one epoch and running one hundred steps with ten epoch? Which one is better? Is there any logic changes between consecutive epochs? Data shuffling?

Answer 1

Steps: A training step is defined as one gradient update

Epoch: An epoch is defined as one full iteration through all the training data.

steps = (epoch * examples)/batch size

For example-

If epoch = 200, examples = 2000 and batchsize = 2000

steps = 200

visit here to know more about TensorFlow Tutorial.