dask / dask-image / 112

100%

main: 99%

push

travis-ci

Refactor out `_unravel_index` function (#60)

* Add a function unravel indices in Dask Arrays

Provides a utility function in `ndmeasure` to handle unraveling a Dask
Array of 1-D indices into multiple dimensions. Uses a simple kernel
function to perform the unraveling on 1-D NumPy Array chunks and coerce
them into a format that Dask can handle. This kernel function is then
used in a Dask Array `map_blocks` call to handle conversion of the full
Dask Array of 1-D indices into N-D indices.

* Test the unravel indices utility function

Make sure that our unravel indices utility function behaves roughly like
NumPy's. They are not a 1-to-1 match as we prefer to return a single
Dask Array instead of a `tuple` of Dask Arrays, which is better suited
for our use case.

* Use the unravel indices utility function

Instead of handling the unraveling of a Dask Array of indices through
mathematical operations in Dask (which creates a more complex graph),
Simply use the utility function, which simply uses `map_blocks` on each
chunk to get the N-D indices simply. This makes it easier for Dask to
potentially fuse this task with other tasks. Also keeps the Dask graph
pretty clean.

200 of 200 branches covered (100.0%)

Branch coverage included in aggregate %.

698 of 698 relevant lines covered (100.0%)

2.99 hits per line