A LSTM layer is a recurrent neural network that uses previous data to predict (more or less). My question comes in order to train it. Due to its use in a time series, it is better to use model.fit()
to fit all at once (suitable in my PC) or train_on_batch()
to upgrade the gradient simulating a time series?
Asked
Active
Viewed 107 times
0

Omar
- 1,029
- 2
- 13
- 33
-
1There is no difference, because each sample in your batch is one time series. – Dr. Snoopy Mar 09 '22 at 21:33
-
@Dr.Snoopy so there is no difference in fitting the whole dataset at once and using `train_on_batch()` with each sample in an incremental learning? – Omar Mar 10 '22 at 07:51
-
Please can you refer [this](https://stackoverflow.com/questions/49100556/what-is-the-use-of-train-on-batch-in-keras) thread may help you. Thanks! – Mar 20 '22 at 12:53