Category Encoding Layer

[AdamLang96]

• Category Encoding Preprocessing Layer Co-authored-by:
  David Kim (@koyykdy) dok098@ucsd.edu
  Brian Zheng (@Brianzheng123) brianzheng345@gmail.com

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

---

This change is 

[AdamLang96]

@mattsoulanille I was hoping to get a review of this whenever you are available to. Thank you!

[mattsoulanille]

Thanks for the PR! I have a few changes for you to take a look at. Feel free to @ me immediately in future PRs, or when you're ready for me to take another look, and I'll respond sooner.

[AdamLang96]

@mattsoulanille Thank you for the review! I'll ping you once these changes have been implemented

[AdamLang96]

Hi @mattsoulanille I made the requested changes and checked that the linter was passing via yarn lint however I am failing the CI build. Do you have any insight on why this could be happening?

[mattsoulanille]

Looks like it's failing one of the bounds tests:

Chrome 105.0.0.0 (Mac OS 10.13.6) Category Encoding webgl2 {"WEBGL_VERSION":2,"WEBGL_CPU_FORWARD":false,"WEBGL_SIZE_UPLOAD_UNIFORM":0} Raises Value Error if max input value !<= numTokens FAILED
	Expected function to throw an exception with message 'Input values must be between 0 < values <= numTokens
	        with numTokens=3', but it threw an exception with message 'Input values must be between 0 < values <= numTokens
	          with numTokens=3'.
	    at <Jasmine>
	    at UserContext.<anonymous> (tfjs-layers/src/layers/preprocessing/category_encoding_test.ts:139:6 <- tfjs/tfjs-layers/src/tfjs-layers_test_bundle.js:55301:59)
	    at <Jasmine>

This is strange; the two error messages look identical to me (unless I've just completely missed something).

Same for output mode tests:

Chrome 105.0.0.0 (Mac OS 10.13.6) Tests for preprocessing utils cpu {} Thows an error if input rank > 2 FAILED
	Expected function to throw an exception with message 'When outputMode is not 'int', maximum output rank is 2
	    Received outputMode 'multiHot' and input shape 2,2,1
	    which would result in output rank 3.', but it threw an exception with message 'When outputMode is not 'int', maximum output rank is 2
	    Received outputMode multiHot and input shape 2,2,1
	    which would result in output rank 3.'.
	    at <Jasmine>
	    at UserContext.<anonymous> (tfjs-layers/src/layers/preprocessing/preprocessing_utils_test.ts:60:6 <- tfjs/tfjs-layers/src/tfjs-layers_test_bundle.js:55403:72)
	    at <Jasmine>

The failures are reproducible locally with yarn bazel run //tfjs-layers:tfjs-layers_webgl2_test (run in the root of the repository). You can focus the test that's failing by replacing it with fit, or you can use this url (gotten by clicking DEBUG in Chrome karma, waiting for the tests to finish, and then clicking the broken one in the list of tests), although you might need to change the port.

Edit: add url

Full logs for layers tests

[mattsoulanille]

Oh, I see the issue. JS does not let you wrap strings by just adding a newline in them. Well, it does, but it includes the newline and all the indentation spaces. You'll need to split those strings into multiple different strings that you add together. I'll add an example.

[mattsoulanille]

Quick review showing how to fix the multiline string errors. I'll do a full review in a bit.

[mattsoulanille]

LGTM with a few nits. Thanks!

[AdamLang96]

@mattsoulanille I just updated the branch and it caused the build to fail. Please let me know if there is something I can fix. Thanks for all of the help!

[mattsoulanille]

@AdamLang96 We just changed some GCP Cloudbuild permissions and they accidentally broke CI for a few minutes. Should be fixed now, so I'll re-run the tests.