Well, it might be a simple question, but I'm just wondering when I should choose batch_input_shape
over input_shape
in Keras LSTM layers. I always use input_shape
and define the batch size at runtime, since most tutorials I've seen so far recommend this method. But then what would be a reason to define the batch size at compile time using batch_input_size
? Which method is preferable and why?
Asked
Active
Viewed 72 times
0

CLRW97
- 470
- 4
- 15
-
Perfect! Thank you! – CLRW97 Nov 09 '21 at 09:46
-
But one more question: Am I right that when using input_shape, batch_size is not fixed and the network can be fed with different batch sizes during runtime and when using batch_input_shape, batch_size is fixed? – CLRW97 Nov 09 '21 at 09:50
-
This depends on the parameter `stateful` of your `LSTM` – AloneTogether Nov 09 '21 at 10:08
-
Does this answer your question? [What's the difference between input\_shape and batch\_input\_shape in LSTM](https://stackoverflow.com/questions/49374508/whats-the-difference-between-input-shape-and-batch-input-shape-in-lstm) – alift Nov 10 '21 at 05:37
-
thank you @alift, but thats the same answer as AloneTogether's – CLRW97 Nov 10 '21 at 08:28