R, keras: How to get output of a hidden layer?

Question

I am using package Keras in R to do a neural network. How may I extract the output from a hidden layer? I found an example in python, but it is just I have no idea how to do that in R.

I'm curious as to why you need to do this? Are you looking to do something with the weights? — I_Play_With_Data, Mar 06 '18 at 13:04
@I_Play_With_Data This is particularly useful when you train an auto-encoder and you want to extract the last layer of the encoder for dimensionality reduction. — David Arenburg, Jan 24 '19 at 09:23

score 8 · Answer 1 · edited Aug 09 '20 at 09:54

8

You can get the answer here.

Here it is:

model <- ...  # create the original model
layer_name <- 'my_layer'
intermediate_layer_model <- keras_model(inputs = model$input,
                                    outputs = get_layer(model, layer_name)$output)
intermediate_output <- predict(intermediate_layer_model, data)

edited Aug 09 '20 at 09:54

Zephyr

997
4
10
20

answered Mar 06 '18 at 02:55

Perochkin

311
2
6

score 0 · Answer 2 · answered Jun 01 '23 at 08:48

I cannot comment as this account is new, so I'll post this as an additional answer:
When using get_layer() using index = ... instead of layer_name = ..., it should be noted that there is a discrepancy between using it as written by Perochkin and using it in the python-style of model$get_layer(...):

model$get_layer(index = as.integer(5))

returns the layer on the zero-based layer index, while

get_layer(model, index = as.integer(5))

or

model %>% get_layer(index = as.integer(5))

is 1-based, so these return different layers. I didn't find this information anywhere else, so I wanted to share, as this cost me quite some time to find out

R, keras: How to get output of a hidden layer?

2 Answers2