[Sync] merge to main for 3.0.2 release - bug fixes; knowledge indexing; future compatibility & dependencies - #3495

Adds a `stacklevel` argument to `deeplabcut.core.deprecation.deprecated()` and uses it when emitting `DLCDeprecationWarning`. This keeps the default behavior (`stacklevel=2`) while allowing wrapper/legacy shim layers to bump the level so warnings point to the real caller.

Replaced direct `plt.cm.get_cmap(...)` calls across visualization and plotting modules with `deeplabcut.utils.matplotlib_future_mode.get_colormap`. This centralizes colormap access, keeps behavior consistent across TensorFlow/PyTorch/GUI/3D paths, and aligns with Matplotlib future-mode compatibility.

Under pandas 3.0 Copy-on-Write, .to_numpy()/.values return read-only
arrays for single-dtype selections; in-place mutation raises
ValueError. Use copy=True at the five affected sites. Pin unchanged.

Follow-up to the pandas 3.0 CoW sweep (#3362). Three more sites extract a
NumPy array from a pandas object and mutate it in place; under pandas 3 CoW
these can be read-only views, raising "assignment destination is read-only".

- refine_training_dataset/tracklets.py: self.data backs self.xy/self.prob,
  which swap_tracklets and the refine GUI mutate in place (HDF load path;
  flagged in review).
- utils/make_labeled_video.py: coords is a view into xyp, masked in place in
  both the first-frame and per-frame branches.
- pose_estimation_tensorflow/.../pose_multianimal_imgaug.py: a single-row
  float Series is read-only; kpts is masked in place when
  mask_kpts_below_thresh is set.

Co-authored-by: Cursor <cursoragent@cursor.com>

Guards against the in-place write on the next line producing a
read-only view when the intermediate frame becomes single-block.

… options

…t of fixed eval_interval=10 for the detector as in base config

…-change

…_names

Previously it duplicated the legacy lookup inline (list(plt.colormaps())) instead of calling the @deprecated-decorated _legacy_get_colormap_names(), so the deprecation warning never fired on old Matplotlib installs, unlike register_colormap/unregister_colormap.

_legacy_register_colormap/_legacy_unregister_colormap/_legacy_get_colormap_names now report their public wrapper names (e.g. register_colormap) in deprecation warnings instead of the private helper name.

update all import sites accordingly

Extend ProjectConfig.normalize_legacy_empty_vales to treat a bare/null YAML value as unset for fields whose type doesn't accept None
(e.g. multianimalproject). This is required to universally support legacy config templates with unset fields.

Mark the parameter as deprecated in utils.read_config. This parameter was only needed for supporting empty values in legacy project configs. Instead, currently a more strict policy is enforced: null/empty is loaded as None, and fails validation if not supported.

since f7305aa None-type support specifically for ProjectConfigs is tolerated using an after validator, which makes the current `ignore_empty` parameter obsolete

DictViewer passes the config path straight to QLabel, which only accepts
str. Since the pathlib migration ConfigEditor hands it a Path, so opening
"Edit config.yaml" raises TypeError. Reproduced on PySide6 6.7.3 and 6.9.3;
it went unnoticed because the GUI tests are skipped where PySide6 is absent.

The deferred reload timer was duplicated in ManageProject and AnalyzeVideos,
and each of the five tabs opened ConfigEditor slightly differently. They now
share DefaultTab._open_config_editor, which keeps the dialog referenced and
parents it to the main window.

That fixes two latent bugs in the pose_cfg / inference_cfg tabs, where the
dialog was a bare local: nothing kept it alive after the method returned, and
without a parent ConfigEditor skipped its screen-relative sizing. Only the
project config asks for a reload, since rebuilding the tabs is wrong for the
other files.

Cover the dialog outliving the call that opened it, the single deferred
reload after saving the project config, the absence of a reload for any other
config, and the no-op when no project is loaded.

Add pre-validation in `plot_evaluation_results` to correctly split and compare ground-truth vs prediction individuals and bodyparts before reshaping arrays. When counts or bodypart sets mismatch, the function now logs a warning and skips that image instead of proceeding with invalid dimensions.

Adds a new test module for `deeplabcut.utils.visualization` focused on evaluation plotting behavior. The tests cover single-animal label handling across scorers, coordinate arrangement for both plotting modes, mismatch/malformed-data skip paths with expected messages, invalid mode validation, output filename conventions in `save_labeled_frame`, and bounding-box linestyle behavior in `make_multianimal_labeled_image`.

Tightened `plot_evaluation_results` by constraining `mode` with a `Literal` type and adding an explicit runtime `ValueError` for unsupported values. The plotting branch logic now treats non-`bodypart` mode consistently as `individual`, and warning messages for individual-count/reshape issues were rewritten to be clearer and more specific. Tests were updated to match the new warning text, and a matplotlib cleanup fixture was added to close figures after each test.

Expose an `overwrite` option in the compatibility `analyze_videos` wrapper, document its PyTorch-only behavior, and pass it through so callers can re-run analysis even when prediction files already exist.

To avoid silently re-running on stale outputs when overwrite is true

This change adds a regression test to ensure the compatibility wrapper passes the `overwrite` flag through to the PyTorch `analyze_videos` API. It fixes the bug where the flag was effectively pinned to `False` when using `Engine.PYTORCH`.

* Fix NumPy torch snapshot loading

Add NumPy 2.0 compatibility for PyTorch snapshot loading by registering the correct scalar reconstruction helper and keeping NumPy float metrics loadable without changing weights_only. If registration fails, emit a warning explaining the fallback and the impact on snapshot loading.

* Ensure affinity stays a float

Add explicit float typing and return values for `Assembly.affinity`, and cast the affinity value before storing it in PyTorch pose tensors. This keeps the inference path type-consistent and avoids integer zero values when no links are present.

* Clarify NumPy safe globals workaround

Expand the snapshot-loading helper docstring to explain why NumPy reconstructors must be allowlisted for legacy metrics, note the NumPy version split more clearly, and mention that newer Torch releases may make the workaround unnecessary.

* Raise numby, add opencv

- opencv was used but never declared as a dependency
- numba .054 is not compatible with our numpy lower bound now

* Update uv.lock

* Add regression tests for Assembly.affinity type

* Fix numba minimum version spec

Correct the `numba` dependency in `pyproject.toml` from `>=0.6` to `>=0.60` so the declared minimum version matches the intended release series.

* Update uv.lock

* Test tweak: avoid typecasting

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Remove redundant float typecast

Co-authored-by: Jaap de Ruyter van Steveninck <32810691+deruyter92@users.noreply.github.com>

---------

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Jaap de Ruyter van Steveninck <32810691+deruyter92@users.noreply.github.com>

Add a `strict_cells` workflow input and use it to make only `matrix.strict` canary lanes fail the job outright. The intelligent testing workflow now enables strict cells for pushes and same-repo PRs while keeping fork PRs non-blocking, and the reporting step skips annotations for lanes that already fail normally.

* release upper bounds for numpy and matplotlib

* keep np upper bound for tensorflow extras

* add numpy-1 to canary lane

* Fix intelligent testing matrix JSON

Add the missing comma between the `matplotlib` and `numpy-1` matrix entries in `.github/workflows/intelligent-testing.yml` so the GitHub Actions matrix parses correctly.

* Expand intelligent test matrix

Broaden the intelligent-testing workflow to include strict dependency-resolution checks for Python 3.10 and 3.12, plus NumPy 1/2 and Matplotlib upgrade variants. This improves compatibility coverage and validates resolver behavior beyond the previous narrower matrix.

* Adjust matplotlib and TF numpy constraints

Restore exclusion of matplotlib 3.7 and 3.7.1 from the base dependencies, and drop the redundant numpy cap from the `tf-cu12` extra because TensorFlow 2.18 already declares the required numpy range.

* Update uv.lock

* Fix missing ceiling in tf

* Update uv.lock

* update pyproject.toml and uv.lock. Only declare np upperbound when required

* Restore numba/opencv floors from #3508 after restack

---------

Co-authored-by: C-Achard <cyril.achard@epfl.ch>

…ments

Follow-up hardening for the LLM knowledge index

Fix hard-coded overwrite kwarg in pytorch `analyze_videos`