matlab/sim​ulink+tens​orflow problem

Hi @Sato, This one's actually more of a MATLAB-Python environment issue than a TensorFlow problem, even though the error makes it look like TF is the culprit. What's happening is that MATLAB launches Python without fully inheriting your conda environment's variables, so TensorFlow's native C++ DLLs can't be located at load time — even though everything looks fine when you run Python directly. First thing to check — run pyenv in your MATLAB command window and see what path it's pointing to. If it's not your matlab_test conda environment, that's likely where things are going wrong. You can point it to the right place with: pyenv(Version='C:\Users\LENOVO\.conda\envs\matlab_test\python.exe') Restart MATLAB after doing this, since pyenv changes don't take effect in the same session. Weblink: https://www.mathworks.com/help/matlab/ref/pyenv.html If pyenv is already pointing to the right env, the next thing to try is switching to OutOfProcess execution mode. By default MATLAB runs Python in the same memory space as itself, and TensorFlow's runtime is quite sensitive to that setup. Switching to OutOfProcess tells MATLAB to spin up a separate Python host process instead, which avoids the DLL conflict entirely: pyenv(ExecutionMode='OutOfProcess') Try your import again after that — this tends to be the fix for most people hitting this exact error. Weblink: https://www.mathworks.com/help/matlab/matlab_external/install-supported-python-implementation.html Also worth thinking about how you're opening MATLAB. If you're launching it from a desktop shortcut or the Start Menu, it won't inherit your conda environment's PATH even if pyenv is configured correctly. Instead, try opening Anaconda Prompt, activating your environment with conda activate matlab_test, and then launching MATLAB from there. That way MATLAB starts with all the right environment variables already in place. Weblink: https://www.mathworks.com/matlabcentral/answers/1450474-how-to-add-the-path-to-a-conda-python-environment One more quick thing to check — run getenv('PYTHONHOME') in MATLAB. If that variable is set and pointing somewhere other than your conda env, it can silently override pyenv and cause exactly this kind of DLL failure. Clearing it with setenv('PYTHONHOME', '') is worth a shot if it's pointing somewhere unexpected. And if none of the above fully resolves it, it's worth knowing that TF 2.19 quietly dropped Python 3.9 support in some Windows builds. As a last resort, try downgrading to tensorflow-cpu 2.15 or 2.16 — both are well tested with Python 3.9 on Windows and might just clear things up. Hope that helps — let me know how it goes!