Installation Issue

[alex-stanoyevitch]

This is my first time putting up an issue in GitHub, so I apologize in advance in case I may be doing something improperly in this submission.

I've been reading the 2nd edition of Chollet's book for a while now and REALLY hope to start using keras + tensorflow.
I have done following the installation instructions in Chap 3 for my PC:

1. First installed Python, Git, and Rtools.
2. Then I ran the following codes provided kindly to me by one of the book's authors (Tomasz K)
   install.packages("keras3")
   reticulate::install_python("3.11")
   keras3::install_keras()

When I paste (in order) all of the RCode snippets from Chap 2 to build the first network, all seems to work fine till I run the snippet:

model <- keras_model_sequential(list(
layer_dense(units = 512, activation = "relu"),
layer_dense(units = 10, activation = "softmax")
))

I get this error message:
'''Error in keras_model_sequential(list(layer_dense(units = 512, activation = "relu"), :
layers must be passed in as a named argument.'''

I'd very much appreciate anyone's help or advice on this.

My background: I am a professor in math and statistics and have been teaching machine learning every year for 8 years and have been using R for 13 years. I've never used Python nor needed to install it till I installed keras3. I am planning a new course in deep learning, but I need to work out such bugs and issues so as to avoid my students running into them. I need a powerful software but also a stable working environment.
My email address is alex.stanoyevitch@gmail.com in case you need or would prefer to contact me directly on this issue.

[t-kalinowski]

Hi, thanks for reporting.

This error occurs if you are using keras instead of keras3.

Can you please try the development version of keras3 and let me know if this works for you?

pak::pak("rstudio/keras3")
print(keras3::op_convert_to_tensor("Hello World!"))
print(reticulate::py_config())

[alex-stanoyevitch]

Thank you Tomasz, I ran these three commands in RStudio, it gave the following output:

pak::pak("rstudio/keras3")
'''! Using bundled GitHub PAT. Please add your own PAT using gitcreds::gitcreds_set().
✔ Updated metadata database: 5.67 MB in 15 files.
✔ Updating metadata database ... done

→ Will update 2 packages.
→ Will download 1 CRAN package (2.26 MB).
→ Will download 1 package with unknown size.

• keras3 1.2.0 → 1.3.0 [bld][cmp][dl] (GitHub: e357997)
• reticulate 1.39.0 → 1.41.0 [dl] (2.26 MB)

! keras3 and reticulate are loaded in the current R session, you probably need to restart R after the installation.

? Do you want to continue (Y/n) print(keras3::op_convert_to_tensor("Hello World!"))
Error: Aborted.

y
[1] -1.0 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0'''

I restarted RStudio and tried running again the same code from Chap 2 of the Chollet book, but when I got to the same snippet in my original post:

model <- keras_model_sequential(list(
layer_dense(units = 512, activation = "relu"),
layer_dense(units = 10, activation = "softmax")
))

But I am still getting the same error.

Do you have anything else that I can try?

[alex-stanoyevitch]

Added information to last post:
I checked the list of installed packages in RStudio; keras3 is there but not keras.

[t-kalinowski]

What is the output of py_config()?

[alex-stanoyevitch]

py_config()
'''Error in py_config() : could not find function "py_config"'''

[t-kalinowski]

reticulate::py_config()

[alex-stanoyevitch]

reticulate::py_config()
'''python: C:/Users/Alex/Documents/.virtualenvs/r-reticulate/Scripts/python.exe
libpython: C:/Users/Alex/AppData/Local/Programs/Python/Python313/python313.dll
pythonhome: C:/Users/Alex/Documents/.virtualenvs/r-reticulate
version: 3.13.1 (tags/v3.13.1:0671451, Dec 3 2024, 19:06:28) [MSC v.1942 64 bit (AMD64)]
Architecture: 64bit
numpy: C:/Users/Alex/Documents/.virtualenvs/r-reticulate/Lib/site-packages/numpy
numpy_version: 2.2.0'''

[t-kalinowski]

Can you please run:

reticulate::virtualenv_remove("r-reticulate")

Alternatively, you can also use the unexported rm_all_reticulate_state to completely revert to a clean slate:

reticulate:::rm_all_reticulate_state(TRUE)

Then try to use keras again.

[alex-stanoyevitch]

I just ran the first option, and tried to build the keras3 model again, but I am getting the same error mssg:

reticulate::virtualenv_remove("r-reticulate")
'''Remove virtual environment 'r-reticulate'? [Y/n]: y
Virtual environment "r-reticulate" removed.'''
model = keras_model_sequential(list(

• layer_dense(units = 512, activation = "relu"),
• layer_dense(units = 10, activation = "softmax")
• ))
  '''Error in keras_model_sequential(list(layer_dense(units = 512, activation = "relu"), :
  could not find function "keras_model_sequential"'''

[alex-stanoyevitch]

Oh I apologize, the keras3 error message now (above) is different.
Should I reinstall keras3 + tensorflow?

[t-kalinowski]

No need to reinstall the R package. It looks like you need to restart your R session and call library(keras3)

[alex-stanoyevitch]

Thank you Tomasz for all of your prompt follow-ups.
I need to leave for work now and also since today is my birthday (I'll be celebrating after work) so will get back to this tomorrow and let you know.

[t-kalinowski]

Happy Birthday!

[alex-stanoyevitch]

Good Morning Tomasz, I have closed all my running RStudio sessions/projects and restarted the one where I am trying to make keras3 work. I redid the library command for keras3 (as you suggested) + for tensor flow. The same code snippet (the model creation) gave an error mssg that was longer than before but the last part was the same as before; here is all the i/o:

library(keras3); library(tensorflow)

'''Attaching package: ‘tensorflow’

The following objects are masked from ‘package:keras3’:

set_random_seed, shape'''

model = keras_model_sequential(list(

• layer_dense(units = 512, activation = "relu"),
• layer_dense(units = 10, activation = "softmax")
• ))
  '''2025-03-05 06:56:36.429873: I tensorflow/core/util/port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
  2025-03-05 06:56:38.208961: I tensorflow/core/util/port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
  Error in keras_model_sequential(list(layer_dense(units = 512, activation = "relu"), :
  layers must be passed in as a named argument.'''

[t-kalinowski]

Hi, I can't reproduce the error locally.

Can you use reprex::reprex() to provide a reproducible example?
https://reprex.tidyverse.org/

Also, please make sure you're running the latest version of the R packages, and to help keep things reproducible, make sure you're using a reticulate-managed python installation.

remotes::install_github("rstudio/reticulate")
remotes::install_github("rstudio/keras3")

Sys.setenv("RETICULATE_PYTHON"="managed")