keras input shape for multivariate LSTM

Question

I'm trying to fit a LSTM model in keras where I have two inputs

y is the output with shape (100,10) x is the input with shape (100,20)

library(keras)

x_train_vec <- matrix(rnorm(2000), ncol = 20, nrow = 100)
x_train_arr <- array(data = x_train_vec, dim = c(nrow(x_train_vec), 1, 20))


y_train_vec <- matrix(rnorm(1000), ncol = 10, nrow = 100)
y_train_arr <- array(data = y_train_vec, dim = c(nrow(x_train_vec), 1, 10))


> dim(x_train_arr)
[1] 100   1  20
> dim(y_train_arr)
[1] 100   1  10

Now I want to fit the LSTM model

model <- keras_model_sequential()

model %>%
  layer_lstm(units            = 50, 
             input_shape      = c(1,10), 
             batch_size       = 1) %>% 
  layer_dense(units = 1)

model %>% 
  compile(loss = 'mae', optimizer = 'adam')

model %>% fit(x          = x_train_arr, 
              y          = y_train_arr, 
              batch_size = 1,
              epochs     = 10, 
              verbose    = 1, 
              shuffle    = FALSE)

But I get this error:

Error in py_call_impl(callable, dots$args, dots$keywords) :
ValueError: Error when checking input: expected lstm_21_input to have shape (1, 10) but got array with shape (1, 20)

If I change input size to c(1,20), I get:

Error in py_call_impl(callable, dots$args, dots$keywords) :
ValueError: Error when checking target: expected dense_13 to have 2 dimensions, but got array with shape (100, 1, 10)

I also played with different setting but it never works.

Answer 1

IF your Keras version is < 2.0 you need to use model.add(TimeDistributed(Dense(1))).

NOTE that syntax is for python, you need to find the R equivealent.

Answer 2

I figured out how to make it work:

x_train_vec <- matrix(rnorm(2000), ncol = 20, nrow = 100)
x_train_arr <- array(data = x_train_vec, dim = c(nrow(x_train_vec), 20, 1))


y_train_vec <- matrix(rnorm(1000), ncol = 10, nrow = 100)
y_train_arr <- array(data = y_train_vec, dim = c(nrow(x_train_vec), 10))


model <- keras_model_sequential()

model %>%
  layer_lstm(units            = 50, 
             input_shape      = c(20,1), 
             batch_size       = 1) %>% 
  layer_dense(units = 10)

model %>% 
  compile(loss = 'mae', optimizer = 'adam')

model %>% fit(x          = x_train_arr, 
              y          = y_train_arr, 
              batch_size = 1,
              epochs     = 10, 
              verbose    = 1, 
              shuffle    = FALSE)