np.matrix still present in NumPy 2.x despite deprecation: safe to keep using it?

Direct answer

No removal target for np.matrix has been published. It still ships in NumPy 2.x, and the removal decision remains open. You can keep using it now, but the supported path forward is migrating to ndarray plus the @ operator.

Confirmed facts vs. likely reasoning

Confirmed: The class is present in the 2.x series. The deprecation warning introduced around NumPy 1.19 has not been converted into a scheduled removal.

Likely explanation: The project appears to be weighing ecosystem compatibility rather than treating this as a routine cleanup. Because np.matrix is an ndarray subclass, a hard removal would ripple through libraries that expose it in public signatures.

Migration path for mixed code

There is no documented automated migration tool. The supported approach is manual and expression-level:

1. Convert np.matrix instances to plain ndarray.
2. Replace matrix-multiplication uses of * with @.
3. Audit every site where np.matrix and ndarray operands meet, because * dispatches to matrix multiplication only when the left operand is a matrix.

The subclass relationship creates a specific hazard: many NumPy functions return plain ndarray when given a matrix, so an expression can silently shift from matrix semantics to element-wise semantics without an obvious type change at the call site.

2.x behavioral checks

NumPy 2.0 changed scalar promotion and copy semantics. For existing np.matrix code, verify these against your 1.x baseline:

• Matrix-scalar operations and resulting dtypes.
• View vs. copy behavior in slicing and arithmetic.
• Shape and dimensionality in mixed matrix/ndarray expressions.

Scoped verification

python -W error::PendingDeprecationWarning -m pytest

1. Promote PendingDeprecationWarning to errors to inventory all construction and usage sites.
2. Convert a representative module to ndarray plus @, then compare numerical outputs and shapes against the original.
3. Search for matrix-specific attributes—.A, .A1, .H, .I—that vanish after conversion.
4. Review the NumPy 2.0 migration guide for subclass and linear-algebra changes.

Release plans can shift, so re-confirm the current np.matrix status against the latest NumPy release notes before locking your timeline.

One detail that changes the recommendation

Does your library return np.matrix to downstream consumers, or is it internal? If it is public API, an internal-only migration leaves consumers exposed to the same future risk and argues for a coordinated deprecation cycle. If it is internal, you can migrate module-by-module without a breaking release.