AttributeError: `np.NaN` was removed in the NumPy 2.0 release. Use `np.nan` instead. (Python) — how to fix
Traceback (most recent call last):
File "clean.py", line 4, in <module>
df = df.replace("", np.NaN)
^^^^^^
File ".../numpy/__init__.py", line 397, in __getattr__
raise AttributeError(
AttributeError: `np.NaN` was removed in the NumPy 2.0 release. Use `np.nan` instead.
Quick fix: In your code, replace the alias: np.NaN/np.NAN → np.nan, np.Inf/np.Infinity/np.infty → np.inf, np.float_ → np.float64, np.complex_ → np.complex128, np.unicode_ → np.str_, np.string_ → np.bytes_. ruff check --select NPY201 --fix does it automatically. If a third-party library raises it, upgrade that library, or pin numpy<2 until it's fixed.
What it means
NumPy 2.0 (June 2024) cleaned up its main namespace and removed about a hundred redundant aliases. Accessing one now raises an AttributeError that names the replacement. The behaviour is identical after you switch, because the aliases pointed at the same objects.
Check your version:
python -c "import numpy; print(numpy.__version__)"
Fix 1: replace the removed names
| Removed in 2.0 | Use instead |
|---|---|
np.NaN, np.NAN |
np.nan |
np.Inf, np.Infinity, np.infty, np.PINF |
np.inf |
np.NINF |
-np.inf |
np.PZERO / np.NZERO |
0.0 / -0.0 |
np.float_ |
np.float64 |
np.complex_, np.cfloat |
np.complex128 |
np.longfloat |
np.longdouble |
np.singlecomplex |
np.complex64 |
np.unicode_ |
np.str_ |
np.string_ |
np.bytes_ |
np.product, np.cumproduct |
np.prod, np.cumprod |
np.alltrue, np.sometrue |
np.all, np.any |
np.round_ |
np.round |
np.asfarray(x) |
np.asarray(x, dtype=np.float64) |
np.cast[dtype](x) |
np.asarray(x, dtype=dtype) |
np.find_common_type |
np.result_type / np.promote_types |
The plain builtins np.float, np.int, np.bool, np.object were already removed in 1.24. Use float, int, bool, object. Note that np.bool was added back in 2.0 as an alias for np.bool_.
Quick search for removed names in your project:
grep -rnE "np\.(NaN|NAN|Inf|Infinity|infty|NINF|PINF|float_|complex_|unicode_|string_)\b" --include=*.py .
Fix 2: let ruff rewrite it
Ruff has a rule just for this migration:
python -m pip install ruff
ruff check --select NPY201 --fix .
It rewrites removed aliases and flags the cases it can't fix automatically.
Fix 3: the error comes from a library
If the traceback ends inside site-packages/<somelib>/, that library hasn't been updated for NumPy 2. Upgrade it first:
python -m pip install -U <somelib>
If no fixed release exists, pin NumPy 1.x in that environment:
python -m pip install "numpy<2"
Add the pin to requirements.txt (numpy<2) and check back later. Libraries built against NumPy 1.x can also fail at import with a different error, A module that was compiled using NumPy 1.x cannot be run in NumPy 2. The same upgrade-or-pin choice applies.
Fix 4: code that must support both 1.x and 2.x
All the replacements above (np.nan, np.inf, np.float64, np.str_, np.bytes_...) already exist in NumPy 1.x. Switching to them works on both versions, so there's no need for version checks.
pandas users
df.replace("", np.NaN) and df.fillna(np.NaN) are the most common triggers. Use np.nan, or pd.NA for nullable dtypes. pandas itself supports NumPy 2 from pandas 2.2.2.
Last updated 2026-11-15.