dask / dask-image / 151

100%

main: 100%

push

travis-ci-com

GPU support for ndfilters & imread modules (#151)

* Add local threshold methods that laxily return dask arrays

* Add tests for local thresholding

* All local_threshold operations should be float

* Import numpy

* Set module name for threshold_local function

* Fix docstring link formatting

* Try flake8 ignore docstring (url can't be shortened)

* threshold_local input argument type checking

* Tests for ndfilters.threshold_local() functions

* Make compatible with older dask versions

* Add test for invalid method name in threhsold_local()

* Remove if statement that is never run

* Older versions of dask.array.from_array() can't handle list input

* Neater way to handle gaussian sigma value in threshold_local()

* Make it work with dask and python 3.5

* threshold_local 'mean' compatible with python 3.5 and 3.7

* Simplify mean filtering in threshold_local()

* Remove unused import from ndfilters._thresholding

* Trigger CircleCI builds to run again.

* GPU dispatch for convolve

* Make convolve function generic for dispatch

* Add forgotten numpy import

* Fix convolve dispatcher name

* Lazy cupyx import

* Lazy import of cupyx.scipy.ndimage

* Fix up flake8 warnings

* Flake8 noqa added

* Update minimum dask dependency to dask==1.1.1

* Remove reference to closed dask issue 6442

* Dispatch from within functions (#5)

* Remove reference to closed dask issue 6442

* Dispatch internall from convolve function

* Call must include an argument

* Restore *args, **kwargs to function signatures

* Yes, you do need to import both cupy and cupyx

* Dispatch from within the convolve function

* Fix up imports

* PEP8 linting tidy up

* John's suggestions for streamlining dispatch

* Use meta kwarg to fix array chunktype repr

* This shouldn't make a difference, but checking just in case

* Fix up dispatch lazy registration, and output array metadata

* Non-lazy cupy di... (continued)

228 of 228 branches covered (100.0%)

Branch coverage included in aggregate %.

853 of 855 relevant lines covered (99.77%)

2.99 hits per line