Array size doubles when specifying periodic axes

[mfschubert]

There is some unexpected behavior when periodic_axes is specified in the length_violations function. Specifically, the violations shape may not match the shape of the input array. Example:

print(f"Density shape is {density.shape}")
violations = imageruler.length_violation(density, diameter=9, periodic_axes=None)
print(f"Violations shape with no periodic axes is {violations.shape}")
violations = imageruler.length_violation(density, diameter=9, periodic_axes=(0, 1))
print(f"Violations shape with periodic axes is {violations.shape}")

Prints the following:

Density shape is (400, 400)
Violations shape with no periodic axes is (400, 400)
Violations shape with periodic axes is (800, 800)

[ianwilliamson]

Possibly due to the tiling that is applied here:

imageruler/imageruler/imageruler.py

Lines 546 to 552 in d77c0af

	periodic_axes = np.array(periodic_axes)
	reps = (2 if 0 in periodic_axes else 1, 2 if 1 in periodic_axes else 1)
	array = np.tile(array, reps)
	phys_size = np.array(phys_size) * reps
	short_entire_side = min(
	phys_size
	) # shorter side of the entire design region

but not(?) undone before returning the violations

[stevengj]

If you know a bugfix, please feel free to submit a PR.

[mawc2019]

This feature is expected and it did not incur an error. To visualize lengthscale violation of a periodic structure, it does no harm to double this structure along the periodic direction(s).

but not(?) undone before returning the violations

It is not undone.

[ianwilliamson]

This feature is expected and it did not incur an error. To visualize lengthscale violation of a periodic structure, it does no harm to double this structure along the periodic direction(s).

It is not expected because the function's doc string states that the returned quantity "indicates violations in the input image:"

imageruler/imageruler/imageruler.py

Line 371 in d77c0af

An array indicating the solid length scale violations in the input image.

A straightforward interpretation is that the mapping is one-to-one. But, if the output array has a different shape than the input array, it is unclear how a user should interpret the quantities in the output array. No documentation or mapping is provided for how violations may be mapped back to features in the input array.

[stevengj]

Yeah, I don't think it makes sense for the function to silently tile the image.

I agree that tiling the image of a periodic structure can be very helpful for visualization, but the user can do this themselves (or there can be a function provided to do this — it's more flexible to provide additional features by composing multiple functions than by adding lots of options to a single function).

[stevengj]

Closed by #30