Any number of channels

[BadrYoubiIdrissi]

Modified the image parametrization to accept any number of channels.

The following choices were made :

• No decorrelation is done across channels.
• A sigmoid is applied on all channels.

On top of the main channels modification, there is a commit to fix a bug switching width and height.

[review-notebook-app]

Check out this pull request on ReviewNB: https://app.reviewnb.com/tensorflow/lucid/pull/179

You'll be able to see visual diffs and write comments on notebook cells. Powered by ReviewNB.

[coveralls]

Pull Request Test Coverage Report for Build 534

• 12 of 14 (85.71%) changed or added relevant lines in 2 files are covered.
• 27 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+0.2%) to 70.492%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
lucid/optvis/param/images.py	11	13	84.62%

Files with Coverage Reduction	New Missed Lines	%
lucid/misc/channel_reducer.py	13	72.34%
lucid/optvis/objectives.py	14	84.08%

Totals
Change from base Build 524:	0.2%
Covered Lines:	1763
Relevant Lines:	2501

---

💛 - Coveralls

[colah]

Nice! Thank you for adding support for this. :)

[colah]

I'm not sure I understand when I'd use this function in practice. Is it just for the tests? Maybe add a comment. :)

[BadrYoubiIdrissi]

Yes exactly :) It's a funky parametrization to test more than 3 channels

[colah]

Makes sense! Consider adding a brief comment, or even moving into the tests? Otherwise, LGTM!

[BadrYoubiIdrissi]

@colah It's done :) I have moved the arbitrary_channels_to_rgb function to the tests as you suggested alongside with an integration level test. I also added the grayscale_image_rgb to the param package init for direct access. Thank you for reviewing !

[colah]

Nice! Thanks for the PR. :)