Trouble with compiling my keras_model_sequential()

[josephkletzer]

Hey, i am fairly new to keras on R. I am now working through the Deep Learning with R book and in the first couple of chapters there is already a load of Errors for me.
So ive figured out how to install keras and tensorflow, like this:

install.packages("remotes")
remotes::install_github("rstudio/reticulate", force = TRUE)
remotes::install_github(sprintf("rstudio/%s", c("tensorflow", "keras")))
reticulate::miniconda_uninstall() # start with a blank slate
reticulate::install_miniconda()
#keras::install_keras()
#keras::install_keras(method = "conda", conda = "auto")
library(keras)
tensorflow::install_tensorflow(conda = "auto", envname = "r-reticulate", version = "release")
reticulate::use_condaenv(condaenv = "r-reticulate", conda = "auto", required = TRUE)`

Then i can import the mnist dataset normally.
After this i use the keras_model_sequential() function like this:

model1 <- keras_model_sequential(list(
  layer_dense(units = 512, input_shape = c(28, 28), activation = "relu", name = "layer1"),
  layer_dense(units = 10, activation = "softmax", name = "output")
))

When i then try to compile() it the following error code appears:

model1 %>% keras::compile(optimizer = "rmsprop",
        loss = "sparse_categorical_crossentropy",
        metrics = "accuracy")


Error: Fehler in UseMethod("compile") : 
  nicht anwendbare Methode für 'compile' auf Objekt der Klasse "c('keras.models.sequential.Sequential', 'keras.models.model.Model', 'keras.backend.tensorflow.trainer.TensorFlowTrainer', 'keras.trainers.trainer.Trainer', 'keras.layers.layer.Layer', 'keras.backend.tensorflow.layer.TFLayer', 'keras.backend.tensorflow.trackable.KerasAutoTrackable', 'tensorflow.python.trackable.autotrackable.AutoTrackable', 'tensorflow.python.trackable.base.Trackable', 'keras.ops.operation.Operation', 'python.builtin.object')" angewendet

So i figured that the class which is produced by my keras_model_sequential() must be different from keras.engine.training.Model, which would be expected as per documentation.

When i call summary(model1) the following appears:

<Sequential name=sequential, built=True>

When i call class(model1) there are multiple results:

[1] "keras.models.sequential.Sequential"                     
 [2] "keras.models.model.Model"                               
 [3] "keras.backend.tensorflow.trainer.TensorFlowTrainer"     
 [4] "keras.trainers.trainer.Trainer"                         
 [5] "keras.layers.layer.Layer"                               
 [6] "keras.backend.tensorflow.layer.TFLayer"                 
 [7] "keras.backend.tensorflow.trackable.KerasAutoTrackable"  
 [8] "tensorflow.python.trackable.autotrackable.AutoTrackable"
 [9] "tensorflow.python.trackable.base.Trackable"             
[10] "keras.ops.operation.Operation"                          
[11] "python.builtin.object"   

How do i get keras_model_sequential() to produce the correct class? I figure thats my problem.. Or what else could be my problem?

[t-kalinowski]

We no longer recommend installing keras with conda - please run the following commands to install keras (please note the '3' in 'keras3'):

install.packages("keras3")
## or, to install the development version
# remotes::install_github("rstudio/keras")
reticulate::install_python()
keras3::install_keras()

There is no need to call use_condaenv() or use_python() these days. A call like library(keras3) will be sufficient to cause reticulate to select the correct r-keras environment that was created by install_keras(). After keras has initialized, please confirm it is using the correct Python virtualenv "r-keras" by calling reticulate::py_config().

The S3 classes of the keras objects can change between releases, and the most likely cause of the issue is that there is a mismatch between the version of the R package installed, and the version of the underlying keras package.

[josephkletzer]

When i try this method it stops at reticulate::install_python() with the following error:

Fehler in pyenv_bootstrap_windows() : 
  Please install git and ensure it is on your PATH

The Way i am installing it right now is the only thing that has worked so far...

[t-kalinowski]

Can you install git as the error message suggests?

https://git-scm.com/download/win

Alternatively, you can install python manually from www.python.org/downloads. Note that the latest release is still too new for the backends, python 3.11 is the latest that is supported.

[josephkletzer]

So i tried everything, and your way of installing works as well. But now i see that more often than not i get this error when i try to create a model:

Fehler in py_call_impl(callable, call_args$unnamed, call_args$named) : ValueError: Only input tensors may be passed as positional arguments. The following argument value should be passed as a keyword argument: <Sequential name=sequential, built=False> (of type <class 'keras.src.models.sequential.Sequential'>)

This is very frustrating, because its difficult to recreate this error, as sometimes when i install everything the first, or first couple of models work, but when i recreate the same models without changing the code, they suddenly turn up this error.... Have you ever seen this before? Do you have any ideas what could be the cause of this?

[josephkletzer]

This is my Traceback:

`── Python Exception Message ────────────────────────────────
Traceback (most recent call last):
File "C:\Users\josep\OneDrive\DOKUME1\VIRTUA1\r-keras\lib\site-packages\keras\src\utils\traceback_utils.py", line 122, in error_handler
raise e.with_traceback(filtered_tb) from None
File "C:\Users\josep\OneDrive\DOKUME1\VIRTUA1\r-keras\lib\site-packages\keras\src\layers\layer.py", line 753, in call
raise ValueError(
ValueError: Only input tensors may be passed as positional arguments. The following argument value should be passed as a keyword argument: (of type <class 'keras.src.models.sequential.Sequential'>)

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

1. ├─... %>% layer_dense(units = 10, activation = "softmax")
2. ├─keras::layer_dense(., units = 10, activation = "softmax")
3. │ └─keras::create_layer(...)
4. ├─keras::layer_dense(., units = 30, activation = "relu")
5. │ └─keras::create_layer(...)
6. └─keras::layer_dense(., units = 512, activation = "relu")
7. └─keras::create_layer(...)

8. ├─keras:::compose_layer(object, layer)
   

9. └─keras:::compose_layer.default(object, layer)
   

10.   └─reticulate (local) layer(object, ...)
    

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

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

reticulate::py_last_error()$r_trace$full_call
[[1]]
keras_model_sequential() %>% layer_dense(units = 512, activation = "relu") %>%
layer_dense(units = 30, activation = "relu") %>% layer_dense(units = 10,
activation = "softmax")

[[2]]
layer_dense(., units = 10, activation = "softmax")

[[3]]
create_layer(keras$layers$Dense, object, list(units = as.integer(units),
activation = activation, use_bias = use_bias, kernel_initializer = kernel_initializer,
bias_initializer = bias_initializer, kernel_regularizer = kernel_regularizer,
bias_regularizer = bias_regularizer, activity_regularizer = activity_regularizer,
kernel_constraint = kernel_constraint, bias_constraint = bias_constraint,
input_shape = normalize_shape(input_shape), batch_input_shape = normalize_shape(batch_input_shape),
batch_size = as_nullable_integer(batch_size), dtype = dtype,
name = name, trainable = trainable, weights = weights))

[[4]]
layer_dense(., units = 30, activation = "relu")

[[5]]
create_layer(keras$layers$Dense, object, list(units = as.integer(units),
activation = activation, use_bias = use_bias, kernel_initializer = kernel_initializer,
bias_initializer = bias_initializer, kernel_regularizer = kernel_regularizer,
bias_regularizer = bias_regularizer, activity_regularizer = activity_regularizer,
kernel_constraint = kernel_constraint, bias_constraint = bias_constraint,
input_shape = normalize_shape(input_shape), batch_input_shape = normalize_shape(batch_input_shape),
batch_size = as_nullable_integer(batch_size), dtype = dtype,
name = name, trainable = trainable, weights = weights))

[[6]]
layer_dense(., units = 512, activation = "relu")

[[7]]
create_layer(keras$layers$Dense, object, list(units = as.integer(units),
activation = activation, use_bias = use_bias, kernel_initializer = kernel_initializer,
bias_initializer = bias_initializer, kernel_regularizer = kernel_regularizer,
bias_regularizer = bias_regularizer, activity_regularizer = activity_regularizer,
kernel_constraint = kernel_constraint, bias_constraint = bias_constraint,
input_shape = normalize_shape(input_shape), batch_input_shape = normalize_shape(batch_input_shape),
batch_size = as_nullable_integer(batch_size), dtype = dtype,
name = name, trainable = trainable, weights = weights))

[[8]]
compose_layer(object, layer)

[[9]]
compose_layer.default(object, layer)

[[10]]
layer(object, ...)

[[11]]
py_call_impl(callable, call_args$unnamed, call_args$named)`

[t-kalinowski]

This error was fixed with a recent release of {keras}. Can you please make sure you're updated?

Also, for new code today, I'd recommend using library(keras3) instead of keras.

[github-actions]

Automatically closed because there has not been a response for 30 days. When you're ready to work on this further, please comment here and the issue will automatically reopen.