Error in pop_layer() for non sequential models

[MaximilianPi]

I'm not sure if this is intended but pop_layer() does not work for non sequential models.
Here a simple example:

x = matrix(runif(200), ncol = 5, nrow = 40)
y = x %*% matrix(1:5, ncol = 1) +3
input = layer_input(shape = 5L)
output = input %>% 
  layer_dense(input_shape = 5L, activation = "relu",units = 20) %>%
  layer_dense(units = 1, activation = "linear") 


model = keras_model(inputs = input, outputs = output)
model %>% 
  compile(
    loss = "poisson",
    optimizer = "rmsprop",
    metrics = "mean_squared_error"
  )
model %>% 
  fit(
    x = x,
    y = y,
    epochs = 5
  )
model %>% pop_layer()

Error: Python object has no '__getitem__' method

sessionInfo()
R version 3.5.0 (2018-04-23)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)
[1]   tensorflow_1.5.0.9001    keras_2.1.6.9001 

[jjallaire]

@kevinushey The source code for pop_layer() is this:

pop_layer <- function(object) {
  object$pop()
}

This looks suspiciously like something caused by rstudio/reticulate#255

[jjallaire]

I didn't realize you were using the Function API. pop() is actually only supported for Sequential models: keras-team/keras#5096