AttributeError: Module NumPy Has No Attribute typeDict

AttributeError: module 'numpy' has no attribute 'typeDict' means some library on your machine is written for an older NumPy. np.typeDict was deprecated and then removed in NumPy 1.24. Your own code almost certainly did not call it — an outdated package did. The fix is to upgrade that package, or to pin NumPy below 1.24 until you can.

What the error looks like#

Traceback (most recent call last):
  File "analyse.py", line 3, in <module>
    import scipy.io
  File ".../site-packages/somepackage/compat.py", line 41, in <module>
    _types = np.typeDict
             ^^^^^^^^^^^
AttributeError: module 'numpy' has no attribute 'typeDict'

Notice that the failing line is not in your file. It is in a package inside site-packages. That package is the one that needs attention.

Why it happens#

np.typeDict was an old, undocumented alias for np.sctypeDict, a dictionary mapping type names to NumPy types. NumPy deprecated it, warned about it for several releases, and removed it in version 1.24 as part of a general clean-up of legacy aliases.

Libraries that had not been updated in a while — older releases of scipy, pandas, statsmodels, gensim, various scientific packages — then broke the moment someone installed them alongside a current NumPy.

Step 1: find out which package#

python -c "import numpy; print(numpy.__version__)"
pip list | grep -i -E "numpy|scipy|pandas|scikit|statsmodels"

Then read the traceback again. The last frame before the error names the file, and the folder just after site-packages is the package name. That is what you upgrade.

Fix 1: upgrade the offending package (preferred)#

pip install --upgrade scipy          # or whichever package the traceback named

Nearly every actively maintained library fixed this within weeks of the NumPy 1.24 release. Upgrading is the correct fix because it moves you forward rather than freezing an old dependency in place.

If several packages are involved, upgrade them together so pip can resolve a consistent set:

pip install --upgrade numpy scipy pandas scikit-learn

Fix 2: pin NumPy below 1.24 (temporary)#

When the package is unmaintained and you cannot replace it today:

pip install "numpy<1.24"

Record it so the environment is reproducible:

numpy<1.24

Treat this as a note to yourself, not a solution. Pinned old dependencies stop receiving security fixes and eventually conflict with something else you need.

python -m venv .venv
source .venv/bin/activate       # .venv\Scripts\activate on Windows
pip install -r requirements.txt

Fix 3: patch it yourself, if the code is yours#

If the failing line is in code you control, the replacement is direct:

import numpy as np

# Old
types = np.typeDict

# New
types = np.sctypeDict

For a third-party package you cannot upgrade, a compatibility shim imported before the package will get you moving, though it is a last resort:

import numpy as np

if not hasattr(np, "typeDict"):
    np.typeDict = np.sctypeDict

import theoldpackage      # import after the shim

Leave a comment explaining why the shim exists and what would let you delete it. Unexplained monkey-patches are how a temporary fix becomes permanent.

The other removals from the same clean-up#

If you hit one of these, you will probably hit the others. The pattern is the same — NumPy removed aliases for built-in Python types:

The general form of the error message is AttributeError: module 'numpy' has no attribute 'X', so once you recognise one you recognise all of them.

Preventing it next time#

# One environment per project
python -m venv .venv

# Record exactly what works
pip freeze > requirements.txt

# Reproduce it elsewhere
pip install -r requirements.txt

Most “it worked yesterday” dependency errors come from installing packages globally, where one project’s upgrade silently changes another project’s environment.

Where to go next#

• Python virtual environments — the thing that prevents most of these.
• Fixing ModuleNotFoundError for pandas — the other common import failure.
• Why is my Python code not working? — how to read a traceback properly.