ImportError: DLL load failed: The specified procedure could not be found.

Not specified by the original author.

TL;DR: your numpy/scipy/sklearn binaries do not line up (a partial upgrade left mismatched DLLs). Reinstall numpy, scipy, and scikit-learn cleanly, ideally in a fresh environment.

The error

ImportError: DLL load failed: The specified procedure could not be found.

Fix it

1. Create a fresh virtual environment (avoids the mixed state entirely):

   python -m venv fresh_env

   Activate it. Expected: clean prompt with the new env.

2. Install the trio fresh:

   pip install scikit-learn

   Expected: pip pulls matching numpy, scipy, and sklearn wheels.

3. Verify:

   python -c "import sklearn; print(sklearn.__version__)"

   Expected: prints the version, no DLL error.

4. If you must keep the old env, force-reinstall instead:

   pip install --force-reinstall numpy scipy scikit-learn

When this applies

Use this when import sklearn on Windows raises the DLL load failed message.

When it does not apply

If the message says the specified module could not be found (rather than procedure), the cause is usually a missing Visual C++ redistributable. On Linux, DLL errors do not apply; look for the .so message instead.

Compatibility

Windows, any recent scikit-learn with numpy/scipy wheels.

Root cause

sklearn's compiled extensions call into numpy/scipy DLLs. When one of the three is upgraded without the others (or a stray DLL shadows the right one), the loader finds a DLL whose exported procedures do not match, and Windows reports the procedure error at import time.

Edge cases

Antivirus quarantining a DLL mid-install produces the same symptom; reinstall with the AV paused if it recurs. Mixing conda and pip installs in one env is the most common way to get into this state.

GitHub issue scikit-learn/scikit-learn#17983 reports ImportError: DLL load failed: The specified procedure could not be found when importing sklearn on Windows. The verified resolution path: the error comes from mismatched or partially upgraded binary dependencies (numpy/scipy DLLs not lining up with what sklearn was built against), and the fix that worked was a clean reinstall of numpy, scipy, and scikit-learn from fresh wheels, ideally in a fresh virtual environment. Full thread: https://github.com/scikit-learn/scikit-learn/issues/17983