Add Filter.Reduce (general dimension reduction for ImageStack)

[kevinyamauchi]

Summary

As discussed in #1302, this component provides the ability to perform dimension reduction operations to ImageStack using DataArray.reduce(). Currently, max, mean, sum, and a user-provided can be applied across specified dimensions. Closes #1302.

Testing

Testing of sum, mean, and max are performed in starfish/core/image/_filter/test/test_reduce.py

To Do

• Add tests
• Update physical coordinates

[codecov-io]

Codecov Report

Merging #1342 into master will increase coverage by 0.04%.
The diff coverage is 94.87%.

@@            Coverage Diff             @@
##           master    #1342      +/-   ##
==========================================
+ Coverage   89.21%   89.25%   +0.04%     
==========================================
  Files         148      149       +1     
  Lines        5377     5416      +39     
==========================================
+ Hits         4797     4834      +37     
- Misses        580      582       +2

Impacted Files	Coverage Δ
starfish/core/image/_filter/reduce.py	94.87% <94.87%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 618e891...161e5a3. Read the comment docs.

[ttung]

why not accept just Callable?

[dganguli]

i like it. this is the more 'pythonic' way in scientific computing -- will be easier for numpy/pandas users to reason about i think.

[ttung]

@dganguli You like what? The original or my suggestion?

[ttung]

In the absence of an opposition, I think func should be a Callable.

[kevinyamauchi]

I think it's nice to have the string option. I think a lot of users will find it annoying to need to import numpy.amax() and this component to perform a maximum intensity projection.

[ambrosejcarr]

I support the string option, just for ecosystem continuity -- users are used to passing function names that are resolved by getattr, as this formalism is used a lot in scipy.spatial.distance among other packages.

[dganguli]

this is awesome. is the plan to re-write max_proj to call reduce with 'max' or 'np.max'? or are we ditching max_proj?

[kevinyamauchi]

@dganguli , I think @ttung suggested that we keep Filter.max_proj() and ImageStack.max_proj() for a bit for API stability. I would be in favor of removing ImageStack.max_proj() and refactoring Filter.max_proj() to use Filter.reduce(). I think it's worth keeping Filter.max_proj() around because maximum intensity projection is such a common operation. Thoughts?

[kevinyamauchi]

@ttung , I think I've addressed your comments. However, the test is failing during doc generation (see error below) and I'm not sure what it is expecting because it passes flake8

Warning, treated as error:
/home/travis/build/spacetx/starfish/starfish/core/image/_filter/reduce.py:docstring of starfish.image.Filter.Reduce:21:Unexpected indentation.
make[1]: *** [html] Error 2
make[1]: Leaving directory `/home/travis/build/spacetx/starfish/docs'
make: *** [docs-html] Error 2
The command "make install-dev fast" exited with 2.

[kevinyamauchi]

@ttung, I think I have addressed your comments. Would you mind taking another look?

[ttung]

In the absence of an opposition, I think func should be a Callable.