How to get the parameters of keras_model_sequential model?

[danli349]

Hello:

How to get the parameters of keras_model_sequential model?
Thanks

model <- keras_model_sequential() %>%
  layer_embedding(input_dim = 10000, output_dim = 32) %>%
  layer_lstm(units = 32) %>%
  layer_dense(units = 1, activation = "sigmoid")
summary(model)

get_layer(model, "lstm_1")$weights

empty

[t-kalinowski]

You can specify a sequential model's input shape and dtype by passing arguments to keras_model_sequential(). See https://keras3.posit.co/reference/keras_model_sequential.html

[danli349]

@t-kalinowski I see, Thanks a lot

[danli349]

@t-kalinowski

I am sorry, Could you let me know why the kernel and recurrent_kernel shape is ([32, 128])?
The layer_embedding output shape is ([32,]), I think the kernel and recurrent_kernel shape would be ([32, 32]).
Thanks

model <- keras_model_sequential(input_shape = c(500)) %>%
  layer_embedding(input_dim = 10000, output_dim = 32) %>%
  layer_lstm(units = 32) %>%
  layer_dense(units = 1, activation = "sigmoid")
summary(model)

get_layer(model, "lstm")$weights |> sapply(\(w) setNames(list(w$shape), w$name)) 

$kernel
TensorShape([32, 128])

$recurrent_kernel
TensorShape([32, 128])

$bias
TensorShape([128])

[t-kalinowski]

It might help to work though what work the LSTM layer is performing. In the "Deep Learning with R" book there is a "lstm from scratch" section with example code: https://github.com/t-kalinowski/deep-learning-with-R-2nd-edition-code/blob/5d666f93d52446511a8a8e4eb739eba1c0ffd199/ch10.R#L327-L342

[danli349]

@t-kalinowski I will take a look. Thanks