AttributeError: module 'numpy' has no attribute 'float'"I hit this one re-running a five-year-old preprocessing script that used to work without complaint. Nothing in the script had changed, only the NumPy version installed under it had, and that was enough to break a line that had been sitting there quietly since a much older project. Below is the reproduction (tested against NumPy 2.4.4), why it only affects some of the old type aliases and not others, and the actual replacement for each one.
The most direct trigger is using np.float (or np.int, np.object, np.str) anywhere a dtype is expected, something that was completely normal NumPy code for years:
import numpy as np
arr = np.array([1, 2, 3], dtype=np.float)
# AttributeError: module 'numpy' has no attribute 'float'.
# `np.float` was a deprecated alias for the builtin `float`. To avoid this
# error in existing code, use `float` by itself. Doing this will not modify
# any behavior and is safe. If you specifically wanted the numpy scalar
# type, use `np.float64` here.It shows up just as often one step removed from a literal dtype= argument, buried inside a .astype() call on a pandas Series pulled out of a DataFrame:
import pandas as pd
df = pd.DataFrame({"a": [1, 2, 3]})
x = df["a"].values.astype(np.float)
# same AttributeError, same traceback shape, just one more frame from pandasnp.float, np.int, np.bool, np.object, and np.str used to exist purely as convenience aliases pointing at Python's own builtin float, int, bool, object, and str, they added nothing NumPy-specific over just writing the builtin directly. NumPy 1.20 (January 2021) marked them deprecated with a warning, and NumPy 1.24 (December 2022) removed most of them outright. Because DeprecationWarning is silenced by Python's default filters outside __main__, that year-plus warning period passed silently for most codebases, and code that hadn't been touched since before 2021 breaks the moment its environment's NumPy gets upgraded past 1.24, with no warning ever having been seen first.
pip show numpy or numpy.__version__ against whatever version the code was originally written against.Every write-up on this error tends to list the same five names together, but they don't all behave the same way today. Verified directly against NumPy 2.4.4:
| Old alias | Result today | Replacement |
|---|---|---|
np.float | Raises AttributeError | float, or np.float64 for the NumPy scalar type |
np.int | Raises AttributeError | int, or np.int64/np.int32 for a specific width |
np.object | Raises AttributeError | object |
np.str | Raises AttributeError | str, or np.str_ for the NumPy scalar type |
np.bool | Does not raise — resolves to <class 'numpy.bool'> | No change needed; it's now NumPy's own boolean scalar type, not the removed builtin alias |
np.long | Does not raise — resolves to <class 'numpy.int64'> | No change needed |
np.bool is the one that trips people up: it was deprecated in the exact same 1.20 release as the other four, for the exact same reason, but NumPy later repurposed the name instead of deleting it, giving it a real, distinct meaning as NumPy's own boolean type. Code (or a blog post) that treats all five as identically broken will get this one wrong.
Pinning numpy<1.24 makes the error go away without touching any code, but it's a stopgap, not a fix, it locks the project out of every NumPy release since December 2022. The actual fix is a direct find-and-replace, and it's safe by NumPy's own description: swapping in the plain Python builtin changes nothing about behavior.
# before
arr = np.array([1, 2, 3], dtype=np.float)
# after — identical behavior
arr = np.array([1, 2, 3], dtype=float)
# or, if a specific NumPy scalar type was actually intended
arr = np.array([1, 2, 3], dtype=np.float64)A quick way to find every instance across a codebase before deciding which replacement fits each one:
grep -rnE "np\.(float|int|object|str)\b" --include="*.py" .If the traceback's last frame before the AttributeError points into site-packages/ rather than a file in the project itself, the removed alias is inside a third-party dependency that hasn't been updated for NumPy 1.24+, not in anything under direct control. Editing the installed package's source is a dead end, it reverts on the next pip install. Two real options in that case:
numpy<1.24 as a deliberate, documented workaround until the dependency catches up, not as the permanent answer.