Installation Issue: Keras3

[Amaizescience]

1. When I tried to download keras3::install_keras, most files downloaded. But there were some errors (See the error message below), and I noticed that a r-tensorflow was not created in the document folder.
   Do you have any suggestions to solve this? (I have most recent version of Microsoft Visual C++ Redistributable for Visual Studio. Also, I checked that there is no python.exe in App execution aliases.)

keras3::install_keras(
  envname = "r-keras",
  python_version = "3.10",
  backend = "tensorflow",
  extra_packages = c("scipy", "pandas", "Pillow", "pydot", "tensorflow_datasets"  ))

_**__**Warning messages:
  1: In file.info(x, extra_cols = FALSE) :
  cannot open file 'C:/Users/scibe/AppData/Local/Microsoft/WindowsApps/python3.exe': The file cannot be accessed by the system
2: In file.info(x, extra_cols = FALSE) :
  cannot open file 'C:/Users/scibe/AppData/Local/Microsoft/WindowsApps/python.exe': The file cannot be accessed by the system**__**_

2. I also tried the recommended installation method: Many installation issues are resolved by running the following in a fresh R session (you can restart R in Rstudio with Ctrl+Shift+F10) :

install.packages("remotes")
remotes::install_github(sprintf("rstudio/%s", c("reticulate", "tensorflow", "keras")))

**_**Warning messages: 
warning messages:
1: In missing_devel_warning(pkgdir) :
  Package reticulate has compiled code, but no suitable compiler(s) were found. Installation will likely fail.
  Install Rtools (https://cran.r-project.org/bin/windows/Rtools/).Then use the pkgbuild package, or make sure that Rtools in the PATH.
2: In i.p(...) :
  installation of package ‘C:/Users/scibe/AppData/Local/Temp/RtmpCMOgLO/file4f1013bf1d50/reticulate_1.44.1.9000.tar.gz’ had non-zero exit status_**


if (is.null(reticulate::virtualenv_starter()))
  reticulate::install_python()
keras3::install_keras()

**_**Warning messages:
1: In file.info(x, extra_cols = FALSE) :
  cannot open file 'C:/Users/scibe/AppData/Local/Microsoft/WindowsApps/python3.exe': The file cannot be accessed by the system
2: In file.info(x, extra_cols = FALSE) :
  cannot open file 'C:/Users/scibe/AppData/Local/Microsoft/WindowsApps/python.exe': The file cannot be accessed by the system_**

[t-kalinowski]

Welcome! Generally, since py_require() was introduced to reticulate, explicitly creating a venv with keras (i.e., calling install_keras()) is no longer necessary.

You can undo any previous steps or actions by calling

reticulate:::rm_all_reticulate_state(TRUE)

Then, you can install the latest version of keras3 from CRAN or github, and everything should "just work".

Install the R package:

install.packages("keras3")
# or 
# remotes::install_github("rstudio/keras3")

Then, restart the session and try using keras. It should "just work" (It might take a few moments the first time to download the dependencies, but subsequent runs should be pretty fast):

keras3::op_convert_to_tensor("Hello World!")

If you encounter any issues or errors, please report back!

[Amaizescience]

Thank you for your reply!
Unfortunately, the error still appears. When I ran the code again, I received the following warning message:

keras3::op_convert_to_tensor("Hello World!")
Downloading uv...Done!
Downloading cpython-3.12.12-windows-x86_64-none (download) (20.8MiB)
 Downloaded cpython-3.12.12-windows-x86_64-none (download)
Downloading numpy (14.8MiB)
Downloading grpcio (4.5MiB)
Downloading tensorboard (5.3MiB)
Downloading setuptools (1.1MiB)
Downloading pandas (10.5MiB)
Downloading pillow (6.7MiB)
Downloading h5py (2.7MiB)
Downloading pygments (1.2MiB)
Downloading jedi (1.5MiB)
Downloading keras (1.4MiB)
Downloading libclang (25.2MiB)
Downloading tensorflow (316.5MiB)
Downloading scipy (36.8MiB)
 Downloaded pygments
 Downloaded h5py
 Downloaded setuptools
 Downloaded keras
 Downloaded grpcio
 Downloaded tensorboard
 Downloaded pillow
 Downloaded pandas
 Downloaded numpy
 Downloaded jedi
 Downloaded libclang
 Downloaded scipy
 Downloaded tensorflow
Installed 59 packages in 17.33s
Error: ImportError: Traceback (most recent call last):
  File "C:\Users\scibe\AppData\Local\R\cache\R\RETICU~1\uv\cache\ARCHIV~1\JHV8YY~1\Lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 73, in <module>
<...truncated...>  module = hook()
             ^^^^^^
  File "C:\Users\scibe\AppData\Local\R\win-library\4.4\reticulate\python\rpytools\loader.py", line 120, in _hook
    return _find_and_load(name, import_)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ImportError: DLL load failed while importing _pywrap_tensorflow_internal: A dynamic link library (DLL) initialization routine failed.


Failed to load the native TensorFlow runtime.
See https://www.tensorflow.org/install/errors for some common causes and solutions.
If you need help, create an issue at https://github.com/tensorflow/tensorflow/issues and include the entire stack trace above this error message.

When I ran the code again, the warning message was:

keras3::op_convert_to_tensor("Hello World!")
List of 23
 $ python              : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/cache/archive-v0/jHV8yy7SlBICAP1fn3piM/Scripts/python.exe"
 $ libpython           : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/python/cpython-3.12.12-windows-x86_64-none/python312.dll"
 $ pythonhome          : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/cache/archive-v0/jHV8yy7SlBICAP1fn3piM"
 $ pythonpath          : chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\reticulate\\uv\\python\\cpython-3.12.12-windows-x86_64-none\\pyt"| __truncated__
 $ prefix              : chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\RETICU~1\\uv\\cache\\ARCHIV~1\\JHV8YY~1"
 $ exec_prefix         : chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\RETICU~1\\uv\\cache\\ARCHIV~1\\JHV8YY~1"
 $ base_exec_prefix    : chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\reticulate\\uv\\python\\cpython-3.12.12-windows-x86_64-none"
 $ virtualenv          : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/cache/archive-v0/jHV8yy7SlBICAP1fn3piM"
 $ virtualenv_activate : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/cache/archive-v0/jHV8yy7SlBICAP1fn3piM/Scripts/activate_this.py"
 $ executable          : chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\RETICU~1\\uv\\cache\\ARCHIV~1\\JHV8YY~1\\Scripts\\python.exe"
 $ base_executable     : chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\reticulate\\uv\\python\\cpython-3.12.12-windows-x86_64-none\\python.exe"
 $ version_string      : chr "3.12.12 (main, Dec  9 2025, 19:02:55) [MSC v.1944 64 bit (AMD64)]"
 $ version             :Classes 'package_version', 'numeric_version'  hidden list of 1
  ..$ : int [1:2] 3 12
 $ architecture        : chr "64bit"
 $ anaconda            : logi FALSE
 $ conda               : chr "False"
 $ numpy               :List of 2
  ..$ path   : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/cache/archive-v0/jHV8yy7SlBICAP1fn3piM/Lib/site-packages/numpy"
  ..$ version:Class 'numeric_version'  hidden list of 1
  .. ..$ : int [1:3] 1 26 4
 $ required_module     : chr "keras"
 $ required_module_path: chr "C:\\Users\\scibe\\AppData\\Local\\R\\cache\\R\\RETICU~1\\uv\\cache\\ARCHIV~1\\JHV8YY~1\\Lib\\site-packages\\keras"
 $ available           : logi TRUE
 $ python_versions     : chr "C:/Users/scibe/AppData/Local/R/cache/R/reticulate/uv/cache/archive-v0/jHV8yy7SlBICAP1fn3piM/Scripts/python.exe"
 $ forced              : chr "py_require()"
 $ ephemeral           : logi TRUE
 - attr(*, "class")= chr "py_config"
Error: Python module keras.ops was not found.

Detected Python configuration:

Do you think this issue might be happening because I installed Python through reticulate?
Would it be better to uninstall that version of Python and install Python manually instead?
Or maybe should I downgrade the version of Python (to 3.10)?

Thank you again for your help!

[t-kalinowski]

I suspect this is a Windows-specific issue that occasionally pops up when the Intel builds of TensorFlow are out of sync with the official TensorFlow builds. I’m not currently on Windows, but you can try setting exclude_newer to a date a few days or months ago (e.g., py_require(exclude_newer = "2025-06-01")) and see if rolling back the clock resolves the issue.

[Amaizescience]

As you mentioned, i guess this is a windows-specific issue.

keras3::op_convert_to_tensor("Hello World!")
Error in system2(uv, c("python list", "--all-versions", "--color never",  : 
  '"C:\Users\scibe\AppData\Local\R\cache\R\reticulate\uv\bin\uv.exe"' not found
Error: Installation of Python not found, Python bindings not loaded.
See the Python "Order of Discovery" here: https://rstudio.github.io/reticulate/articles/versions.html#order-of-discovery.

But I appreciate your help and the time you spent troubleshooting this with me.
Thanks!

[t-kalinowski]

'"C:\Users\scibe\AppData\Local\R\cache\R\reticulate\uv\bin\uv.exe"' not found

I think this particular error was fixed in the latest release of reticulate. Can you confirm if you're using the latest reticulate with install.packages("reticulate")?

[Amaizescience]

Yes, its version is 1.44.1
1......................

packageVersion("reticulate")
[1] ‘1.44.1’

2..................

install.packages("keras3")
WARNING: Rtools is required to build R packages but is not currently installed. Please download and install the appropriate version of Rtools before proceeding:

https://cran.rstudio.com/bin/windows/Rtools/
Installing package into ‘C:/Users/scibe/AppData/Local/R/win-library/4.4’
(as ‘lib’ is unspecified)
trying URL 'https://cran.rstudio.com/bin/windows/contrib/4.4/keras3_1.4.0.zip'
Content type 'application/zip' length 13137547 bytes (12.5 MB)
downloaded 12.5 MB

package ‘keras3’ successfully unpacked and MD5 sums checked

The downloaded binary packages are in
	C:\Users\scibe\AppData\Local\Temp\RtmpgXR6QS\downloaded_packages

3.........................

keras3::op_convert_to_tensor("Hello World!")
Downloading uv...Done!
Downloading cpython-3.12.12-windows-x86_64-none (download) (20.8MiB)
 Downloaded cpython-3.12.12-windows-x86_64-none (download)
Downloading libclang (25.2MiB)
Downloading setuptools (1.1MiB)
Downloading jedi (1.5MiB)
Downloading pandas (10.5MiB)
Downloading h5py (2.7MiB)
Downloading scipy (36.8MiB)
Downloading pillow (6.7MiB)
Downloading grpcio (4.5MiB)
Downloading pygments (1.2MiB)
Downloading numpy (14.8MiB)
Downloading tensorflow (316.5MiB)
Downloading tensorboard (5.3MiB)
Downloading keras (1.4MiB)
 Downloaded pygments
 Downloaded h5py
 Downloaded setuptools
 Downloaded keras
 Downloaded grpcio
 Downloaded tensorboard
 Downloaded jedi
 Downloaded pillow
 Downloaded pandas
 Downloaded numpy
 Downloaded libclang
 Downloaded scipy
 Downloaded tensorflow
Installed 59 packages in 22.01s
Error: ImportError: Traceback (most recent call last):
  File "C:\Users\scibe\AppData\Local\R\cache\R\RETICU~1\uv\cache\ARCHIV~1\T145CV~1\Lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 73, in <module>
<...truncated...>  module = hook()
             ^^^^^^
  File "C:\Users\scibe\AppData\Local\R\win-library\4.4\reticulate\python\rpytools\loader.py", line 120, in _hook
    return _find_and_load(name, import_)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ImportError: DLL load failed while importing _pywrap_tensorflow_internal: A dynamic link library (DLL) initialization routine failed.


Failed to load the native TensorFlow runtime.
See https://www.tensorflow.org/install/errors for some common causes and solutions.
If you need help, create an issue at https://github.com/tensorflow/tensorflow/issues and include the entire stack trace above this error message.

[t-kalinowski]

I am unable to reproduce the error. I installed keras3, tensorflow, and reticulate from CRAN and did not run into any issues. Screenshots are below.

Since the issue seems to be with TensorFlow specifically, can you try using a different backend—JAX or Torch? Call use_backend("jax") or use_backend("torch") at the start of the R session.

Note that in the screenshots I cleared reticulate’s cache, so there are a number of initial downloads. Those occur on first run only. Subsequent calls should be very fast, with no downloads (last screenshot).

• Fresh state, first run, no issues:

---

• With the JAX backend:

---

• With the Torch backend:

---

Subsequent runs reuse the cache instead of downloading and are fast:

[Amaizescience]

Unfortunately, it’s still not working. I even tried running cmd: pip install keras tensorflow, but no luck.

I’m not sure if the issue is due to my AppData folder being read-only or Documents folder being synced with OneDrive on my wife's laptop. (I had installation problems on my laptop and thought it was due to its age, so I'm trying with her laptop currently)

Fortunately, the older version of Keras is still working. Hopefully, after I buy a new laptop this month, the installation will work properly.