added a Feature Histogram Widget

[jo-mueller]

Linked to issue #58

New features:

This PR introduces a new widget to create Histograms from feature data in Labels/Points/Shapes/Tracks/Vectors layers. It's strongly based off of the ScatterWidget. I wasn't sure whether scatter.py was the right place to put it, but it felt like a better fit than histogram.py, which essentially just works with image intensity data rather than tabular data.

Example code snippet

This code snipped demonstrates what it does:

import numpy as np
import napari
from napari_matplotlib import HistogramWidget, FeaturesHistogramWidget

n_points = 1000
random_points = np.random.random((n_points,3))*10
feature1 = np.random.random(n_points)
feature2 = np.random.normal(size=n_points)

viewer = make_napari_viewer()
viewer.add_points(random_points, properties={'feature1': feature1, 'feature2': feature2}, face_color='feature1', size=1)

widget = FeaturesHistogramWidget(viewer)
viewer.window.add_dock_widget(widget)
widget._set_axis_keys('feature1')
widget._key_selection_widget()
widget._set_axis_keys('feature2')
widget._key_selection_widget()

This code is essentially also what I added to the tests (tests are passing locally).

Looking forward to hear your thoughts on this! :)

[jo-mueller]

PS: @dstansby I didn't see that there was ongoing work on a different PR because it wasn't linked to the issue. Feel free to treat this as stale for while.

[dstansby]

Looking good - a few major changes to make though. I have time to revise this myself, or I can leave you to do it- let me know either way!

[jo-mueller]

Hi @dstansby ,

thanks a lot for taking the time to review this! I think I can definitely get those changes done myself. I'll be on vacation in the coming week so PR should come sometime in the second-to-next week.

[jo-mueller]

Hi @dstansby ,

I added your suggestions from above and replaced the magicgui widgets with pyqt code as has been done in the scatter widget in the meantime. Also sorry for the messy commit history - let me know if I should put the changes on a clean branch and re-start the PR.

[dstansby]

pre-commit.ci autofix

[dstansby]

Looks good! I've left one requst for a change. In addition, could you add a figure test? Hopefully the other test files help with that, but let me know if you need any pointers.

Otherwise I'm 👍 to merge.

[jo-mueller]

Hi @dstansby ,

I'm not sure what you mean by figure test? I did add the following to the tests. This adds two points layers to the viewer with two sets of features assigned to each. I check whether the figure in the widget changes upon changing the key for the histogram and whether the plot changes as it should when the selected layer is changed.

    n_points = 1000
    random_points = np.random.random((n_points, 3)) * 10
    feature1 = np.random.random(n_points)
    feature2 = np.random.normal(size=n_points)

    viewer = make_napari_viewer()
    viewer.add_points(
        random_points,
        properties={"feature1": feature1, "feature2": feature2},
        name="points1",
    )
    viewer.add_points(
        random_points,
        properties={"feature1": feature1, "feature2": feature2},
        name="points2",
    )

    widget = FeaturesHistogramWidget(viewer)
    viewer.window.add_dock_widget(widget)

    # Check whether changing the selected key changes the plot
    widget._set_axis_keys("feature1")
    fig1 = deepcopy(widget.figure)

    widget._set_axis_keys("feature2")
    assert_figures_not_equal(widget.figure, fig1)

    # check whether selecting a different layer produces the same plot
    viewer.layers.selection.clear()
    viewer.layers.selection.add(viewer.layers[1])
    assert_figures_equal(widget.figure, fig1)

[dstansby]

For a figure test, see e.g.,

napari-matplotlib/src/napari_matplotlib/tests/test_histogram.py

Lines 8 to 16 in 87d34f5

	@pytest.mark.mpl_image_compare
	def test_histogram_2D(make_napari_viewer, astronaut_data):
	viewer = make_napari_viewer()
	viewer.theme = "light"
	viewer.add_image(astronaut_data[0], **astronaut_data[1])
	fig = HistogramWidget(viewer).figure
	# Need to return a copy, as original figure is too eagerley garbage
	# collected by the widget
	return deepcopy(fig)

This using the pytest-mpl plugin to compare the generated image with an image stored in the repository in this directory: https://github.com/matplotlib/napari-matplotlib/tree/main/src/napari_matplotlib/tests/baseline

[jo-mueller]

For the figure test: So @pytest.mark.mpl_image_compare will compare the returned image to any image in the baseline directory?

[dstansby]

No, you'll need to generate the new test image - see https://github.com/matplotlib/pytest-mpl#usage for instructions.

[jo-mueller]

@dstansby

No, you'll need to generate the new test image - see matplotlib/pytest-mpl#usage for instructions.

Ok, I added the baseline image, but for some reason, the generated baseline is not detected by the test...Sorry for the back and forth, I'm not so familiar with these testing schemes 🙈

Just as a sidenote: Setting everything up for pytest --mpl seemed quite complicated to me. It requires a few additional installation steps (pip install pyqt6 tinycss2) that did not seem evident to me from the.

[dstansby]

No worries! If you don't mind, I'd be happy to pick this up and get the test working?

[jo-mueller]

No worries! If you don't mind, I'd be happy to pick this up and get the test working?

Not at all! Would be cool though if it were documented somewhere for future contributions 👍

[dstansby]

Looks like the test image was just in the wrong directory :) I also did a few other minor fixes, will merge this and do a new release of napari-matplotlib if the tests pass. Thanks a lot for the contribution, and sorry for how long it took me to merge it...

[dstansby]

Looks like it's just code coverage failing