1
$\begingroup$

What is the logic of the epoch? For example, 1 time, 2 times, etc. I just do not know what else is working to give better results than I know.

$\endgroup$

2 Answers 2

3
$\begingroup$

There are two terms which relate to the number of examples while learning. Epoch and iteration. During each epoch, you feed all the examples in your training set and update the network. You can feed the data simultaneously, not for large-scale tasks, or batch by batch. Each turn you pass each batch is called iteration. Consequently, each epoch may have multiple iterations. The reason we have to pass the data multiple times is that we don't know the height of each local optimum along each axis, feature. Consequently, we use a learning rate to limit our steps to small ones and take multiple steps to get close to the desired point.

Due to the request of one of our friends I add the explicit definitions:

  • Epoch: It means how many times the entire dataset has been passed through the network.
  • Iteration: It means for each epoch, how many times you have passed the chunks of the dataset. These chunks of data are called batch and the number of times you pass them through the network is called the number of iterations.
$\endgroup$
7
  • $\begingroup$ The difference between epoch and iteration definition should be made more readable by suitable formatting $\endgroup$
    – DuttaA
    Aug 4, 2018 at 13:54
  • $\begingroup$ @DuttaA I'm open to criticism, which part you think is not readable? $\endgroup$ Aug 4, 2018 at 14:09
  • $\begingroup$ I have understood the answer, i just think many users who does not have a pre-emptive knowledge of the difference would not understand the difference...like iterations are over batches or examples or epochs...whereas epochs are over entire training set no matter the method....the difference should be stated in bold like : epoch : <definition> iterations <definition> or any form seen suitable $\endgroup$
    – DuttaA
    Aug 4, 2018 at 14:12
  • $\begingroup$ (@media )Thanks, can we say: Epoch: a complete tour.iteration: fragmented, full tour of one. $\endgroup$
    – Developer
    Aug 5, 2018 at 6:30
  • $\begingroup$ @Engin I didn't understand the last part. $\endgroup$ Aug 5, 2018 at 6:51
1
$\begingroup$

In a simple way, Epoch is an iteration. For example, in k-means clustering, in each epoch you get better clusters. enter image description here

After a successful epoch the clusters in (a) are approximated better to clusters in (b).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.