In Keras, How can i get 4D output from LSTM

Asked Nov 14 '17 at 17:49

Active Nov 14 '17 at 17:49

Viewed 388 times

I want to implement a model with LSTM and one pooling layer which wants 4D input. However LSTM's output is 3D. How can i implement ?

I want model

LSTM ( units:32, input_shape(batch_size,1,200),return_sequences:true)
LSTM ( units:32, input_shape(batch_size,1,200),return_sequences:true)  //3D output
pooling() //input_Shape(sample,row,col,channels)
dense()

asked Nov 14 '17 at 17:49

melihbasbug

1

How about `Pooling1D`? – Daniel Möller Nov 14 '17 at 17:54
1

It's output '3D tensor with shape' – melihbasbug Nov 14 '17 at 18:06
1

Pooling1D is for 3D tensors. – Daniel Möller Nov 14 '17 at 18:30
What is your data shape? – Marcin Możejko Nov 14 '17 at 18:46
1

my data shape (data[0],1,data[1]) I changed 2D shape to 3D shape for LSTM... data was (1821,20), now (1821,1,20) – melihbasbug Nov 14 '17 at 20:48
1

If you need to apply a dense after the 2nd LSTM for regression or clasification, have you tried with `return_sequences=False` (in the 2nd LSTM)? That would be the vanilla way to combine LSTMs and dense layers. If this is not what you want/need, please tell us more info on what you need to do the pooling for, thanks! – JARS Nov 15 '17 at 00:23

In Keras, How can i get 4D output from LSTM

0 Answers0