Error in py_call_impl(callable, call_args$unnamed, call_args$named) : ValueError: Input 2 of layer "functional_4" is incompatible with the layer:

[yuriy-babenko]

https://github.com/t-kalinowski/deep-learning-with-R-2nd-edition-code/blob/5d666f93d52446511a8a8e4eb739eba1c0ffd199/ch07.R#L146C1-L150C2

Executed code:

`install.packages("remotes")
remotes::install_github("rstudio/tensorflow")
reticulate::install_python()
install.packages("keras3")
keras3::install_keras(envname = "r-reticulate")
library(keras3)

model <- keras_model_sequential() %>%
layer_dense(units = 64, activation = "relu") %>%
layer_dense(units = 10, activation = "softmax")

Not build yet, not weights

model$weights

model$build(input_shape = shape(NA, 3))
str(model$weights)

model

model <- keras_model_sequential(input_shape = c(3), name = "my model" ) %>%
layer_dense(units = 64, activation = "relu") %>%
layer_dense(units = 10, activation = "softmax")

model

Functional API

inputs <- layer_input(shape = c(3), name = "my_input")
features <- inputs %>% layer_dense(64, activation = "relu")
outputs <- features %>% layer_dense(10, activation = "softmax")
model <- keras_model(inputs = inputs, outputs = outputs)

model

multi-input, multi-output

vocabulary_size <- 10000
num_tags <- 100
num_departments <- 4

title <- layer_input(shape = c(vocabulary_size), name = "title")
text_body <- layer_input(shape = c(vocabulary_size), name = "text_body")
tags <- layer_input(shape = c(vocabulary_size), name = "tags")

features <-
layer_concatenate(list(title, text_body, tags)) %>%
layer_dense(64, activation = "relu")

priority <- features %>%
layer_dense(1, activation = "sigmoid", name = "priority")

department <- features %>%
layer_dense(num_departments, activation = "softmax", name = "departments")

model <- keras_model(
inputs = list(title, text_body, tags),
outputs = list(priority, department)
)

num_samples <- 1280
random_uniform_array <- function(dim)
array(runif(prod(dim)), dim)

random_vectorized_array <- function(dim)
array(sample(0:1, prod(dim), replace = TRUE), dim)

title_data <- random_vectorized_array(c(num_samples, vocabulary_size))
text_body_data <- random_vectorized_array(c(num_samples, vocabulary_size))
tags_data <- random_vectorized_array(c(num_samples, num_tags))

priority_data <- random_vectorized_array(c(num_samples, 1))
department_data <- random_vectorized_array(c(num_samples, num_departments))

model %>% compile(
optimizer = "rmsprop",
loss = c("mean_squared_error", "categorical_crossentropy"),
metrics = c("mean_absolute_error", "accuracy")
)

model %>% fit(
x = list(title_data, text_body_data, tags_data),
y = list(priority_data, department_data),
epochs = 1
)`

Error message:

Error in py_call_impl(callable, call_args$unnamed, call_args$named) : ValueError: Input 2 of layer "functional_4" is incompatible with the layer: expected shape=(None, 10000), found shape=(32, 100) Run reticulate::py_last_error() for details.

Output of detailed log:

`── R Traceback ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
▆

1. ├─model %>% ...
2. ├─generics::fit(...)
3. └─keras3:::fit.keras.src.models.model.Model(...)
4. ├─base::do.call(object$fit, args)
5. └─reticulate (local) <python.builtin.method>(...)

6. └─reticulate:::py_call_impl(callable, call_args$unnamed, call_args$named)
   

See reticulate::py_last_error()$r_trace$full_call for more details.`

If I understand the issue correctly this is related to the move to Keras3 mentioned here: rstudio/keras3#1427 (comment)

Looking at keras3: https://keras.io/api/layers/core_layers/embedding/, there is indeed no "input_length" any longer but what I still do not understand is where do we use embedding in the above code? We are accessing the "Functional API" and nothing in code suggests we call "Embedding Layer".