dask / dask-image / 170

100%

main: 100%

push

travis-ci

Use 1-D structured array fields for position-based kernels in `ndmeasure` (#79)

* Use N-D structured array field in `center_of_mass`

Instead of using multiple structured array fields to pass back the
different coordinate components of the center of mass, use one 1-D field
with the size being the number of components. This simplifies the array
handling a bit and avoids the need to split apart and restack the array.

* Use N-D structured array in `maximum_position`

Rewrite `maximum_position` to use a structured `dtype` with a 1-D field.
This allows the unraveled position to be determined in the
`maximum_position` kernel function and returned. With this change there
is no need to handle unraveling of the position in the Dask graph. Only
the structured field need be selected, which is a constant time
operation. Thus simplifying the Dask graph generated by
`maximum_position` and improving the computation performance as well.

* Use N-D structured array in `minimum_position`

Rewrite `minimum_position` to use a structured `dtype` with a 1-D
field.  This allows the unraveled position to be determined in the
`minimum_position` kernel function and returned. With this change there
is no need to handle unraveling of the position in the Dask graph. Only
the structured field need be selected, which is a constant time
operation. Thus simplifying the Dask graph generated by
`minimum_position` and improving the computation performance as well.

* Use N-D structured array fields in `extrema`

Make use of N-D structured array fields in `extrema`'s kernel function
to pass back N-D coordinate positions. Avoids having an explicit
unraveling step added to the Dask graph after the `extrema`'s kernel
function.

* Drop `_unravel_index` from `ndmeasure._utils`

This function is no longer needed as unraveling of indices now occurs
within kernel functions that are returning position. Hence it is dropped
along with its tests.

209 of 209 branches covered (100.0%)

Branch coverage included in aggregate %.

728 of 728 relevant lines covered (100.0%)

2.99 hits per line