2
$\begingroup$

I am training a char-rnn to yield some nice generative output. Can I make it memorize context by using the same copies of data in training, multiple times?

$\endgroup$
3
$\begingroup$

Repeated use of the same data in fitting just gives that data too much weight. It provides a false indication of the "true" sample size and hence makes the variance of parameter estimates improperly small.

$\endgroup$
2
$\begingroup$

This is another question which addresses the same point: https://stackoverflow.com/questions/38705166/repeating-training-with-the-same-data-set

You are risking over-fitting but if you add noise as suggested in the answer it could help you.

$\endgroup$
2
  • $\begingroup$ Also, is overfitting the same as memorizing the training data. I understand the term overfitting, but intuitively it is actually memorizing data instead of generalizing right? $\endgroup$ – Rafael Nov 29 '16 at 6:37
  • $\begingroup$ Yes, you start learning each case in training, and not generalizing $\endgroup$ – Arun Jose Nov 29 '16 at 6:51

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.