This tracking issue is meant to track the status and tasks of the Python API cleanup project for NumPy 2.0 (NEP 52, currently in draft status). Note: this is currently far from a complete plan, I wanted to make a start though at tracking things and making them actionable.

This issue is probably also a good place to suggest additional APIs to move/remove/deprecate, or other open issues that are related and can be tackled.

As a way of working, it'd be good to check usages of the function/object one is working on in downstream libraries (SciPy, scikit-learn, pandas is a good start), clean those up, then remove the object in question from the API. That has at least two benefits:

1. it avoids CI in those projects to start failing when the change lands in a NumPy nightly,
2. it gives a good sense of how easy it is to replace the usage with the recommended replacement

Cleaning up the main namespace

• Deal with all functions listed in Document, deprecate or remove everything exposed in the "numpy" namespace #12385
• Remove np.inf and np.nan aliases
• Deprecate np.compat: API: deprecate compat and selected lib utils #23830
• See ENH: Overhaul of NumPy main namespace [NEP 52] #24306 for many more TODOs and decisions

Cleaning up the submodule structure

• Clean up numpy.lib: ENH: Overhaul of NumPy lib namespace [NEP 52] #24507

Reducing the number of ways to select dtypes

Actionable:

• Implement np.isdtype
• Make adding new things to sctypeDict generate a deprecation warning. We cannot completely remove it because jax does this. See DEP: Deprecate registering dtype names with np.sctypeDict? #24699.
• Deprecate (in 1.26.0) and remove (in 2.0) sctype related things (see ENH: add a canonical way to determine if dtype is integer, floating point or complex #17325 (comment))
• Remove issubsctype, sctypeDict & co from SciPy, pandas (both have a few occurrences) and perhaps a few other large downstream libraries
• Remove some of the more obscure aliases listed in https://numpy.org/devdocs/reference/arrays.scalars.html#other-aliases.

Cleaning up the niche methods on numpy.ndarray

There are the ones listed in the NEP right now

• .setitem
• .newbyteorder
• .ptp

Doing the above ones will give a better idea about the amount of effort involved, and may help with then identifying a next set.

Documentation

• Update docs for the submodule structure (done in DOC: add a section on NumPy's module structure to the refguide #25674)
• Write a migration guide: https://numpy.org/devdocs/numpy_2_0_migration_guide.html
• Document how to use a tool or script for migrating to the 2.0 API: in progress with ruff by @mtsokol
• Implement a .. legacy:: Sphinx directive - in progress: DOC: Add legacy directive to mark outdated objects #24939
• Mark legacy APIs with .. legacy:: directive