Optimized :meth:.Axes.coords_to_point

[chopan050]

Overview: What does this pull request change?

As the title says, I optimized Axes.coords_to_point. Best results if combined with PR #3285: Optimized :meth:NumberLine.number_to_point.

Motivation and Explanation: Why and how do your changes improve the library?

• Mainly, I got rid of the np.sum and the list comprehension, and replaced them with a good ol' for loop, which is much more efficient in this case because there are only 2 or 3 coords to add.
• I avoided using VMobject.__iter__ on self.axes by iterating directly over its submobjects attribute.
• Also, I removed one unnecessary addition and subtraction of origin by assigning first the result of self.x_axis.number_to_point(coords[0]) by separate (which should not be a problem since x_axis is already being used explicitly to calculate origin.
• Finally, in general I worked to make the code more readable.

The changes I propose:

    def coords_to_point(
        self, *coords: float | Sequence[float] | Sequence[Sequence[float]] | np.ndarray
    ) -> np.ndarray:
        coords = np.asarray(coords)
        origin = self.x_axis.number_to_point(
            self._origin_shift([self.x_axis.x_min, self.x_axis.x_max]),
        )

        # Is every component in coords in the format [xi, yi, zi]?
        # i.e. is coords in the format ([[x1 y1 z1] [x2 y2 z2] ...])? (True)
        #
        # Or is coords in the format (x, y, z) or ([x1 x2 ...], [y1 y2 ...], [z1 z2 ...])? (False)
        #
        # The latter is preferred.
        are_components_xyz = False

        # If coords is in the format ([[x1 y1 z1] [x2 y2 z2] ...]):
        if coords.ndim == 3:
            # Extract from original tuple: now the format is [[x1 y1 z1] [x2 y2 z2]]
            coords = coords[0]

            # If there's a single coord, extract it so that coords_to_point returns a single point
            if coords.shape[0] == 1:
                coords = coords[0]  # In this case, now coords = [x1 y1 z1]
            else:
                are_components_xyz = True
                # Transform coords into the format [[x1 x2 ...] [y1 y2 ...] [z1 z2 ...]]
                # for later processing.
                coords = coords.T

        # Now coords should be in the format [x y z], where each component is either a float or an ndarray
        points = self.x_axis.number_to_point(coords[0])
        other_axes = self.axes.submobjects[1:]
        for axis, nums in zip(other_axes, coords[1:]):
            points += axis.number_to_point(nums) - origin

        if are_components_xyz:
            return points
        return points.T  # Has no effect on a 1D array representing a single point

The results before and after the changes (using the same test scene I used for #3285). As you can see, I managed to cutoff around 0.70 seconds from Axes.coords_to_point. If the changes in the previous PR get accepted as well, there's even more speedup:

Before	After

Links to added or changed documentation pages

Further Information and Comments

Reviewer Checklist

• The PR title is descriptive enough for the changelog, and the PR is labeled correctly
• If applicable: newly added non-private functions and classes have a docstring including a short summary and a PARAMETERS section
• If applicable: newly added functions and classes are tested

[behackl]

One minor suggestion, otherwise this is a fine improvement. Please take a look & thanks for your efforts!

[behackl]

I think this could be made even more explicit: are_components_xyz is not particularly self-explanatory. Why not are_coordinates_transposed?

[chopan050]

Yeah, I like that one better. The original transposed variable was not really clear IMO, but in hindsight are_components_xyz is not much better.

Gonna change that!

[chopan050]

Done! I changed the variable name to are_coordinates_transposed, I liked that one better.

I also added some more commentaries and added an extra condition points.ndim == 1 to the final return part. It bothered me a bit that the "single point" case falled into the return points.T situation and wanted to make more explicit that the single point should also be returned as is, but I forgot to add that before.

[behackl]

LGTM, thank you very much!