[WIP] Add WordCorrelationPlot visualizer

[pdeziel]

This PR will fix #148 which requested a WordCorrelationPlot visualizer to allow the user to visualize the correlations between words in the corpus. I came across this issue in the backlog and I can see myself using it so I'd be happy to work on it.

Sample Code and Plot

TODO

TODOs and questions

Still to do:

• [x] Implement the visualizer
• [x] Create the tests
• [x] Add the documentation

CHECKLIST

• [ ] Is the commit message formatted correctly?
• [ ] Have you noted the new functionality/bugfix in the release notes of the next release?

• [x] Included a sample plot to visually illustrate your changes?
• [x] Do all of your functions and methods have docstrings?
• [x] Have you added/updated unit tests where appropriate?
• [x] Have you updated the baseline images if necessary?
• [x] Have you run the unit tests using pytest?
• [x] Is your code style correct (are you using PEP8, pyflakes)?
• [x] Have you documented your new feature/functionality in the docs?

• [x] Have you built the docs using make html?

[codecov[bot]]

Codecov Report

Merging #1243 (a2b80a5) into develop (7e238e1) will increase coverage by 0.13%. The diff coverage is 98.83%.

@@             Coverage Diff             @@
##           develop    #1243      +/-   ##
===========================================
+ Coverage    90.62%   90.75%   +0.13%     
===========================================
  Files           92       93       +1     
  Lines         5214     5300      +86     
===========================================
+ Hits          4725     4810      +85     
- Misses         489      490       +1     

Impacted Files	Coverage Δ
yellowbrick/text/correlation.py	98.82% <98.82%> (ø)
yellowbrick/text/__init__.py	100.00% <100.00%> (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 7e238e1...a2b80a5. Read the comment docs.

[lwgray]

Thanks for opening a PR. Please move forward with the development of this feature. All the best

[lwgray]

@pdeziel bump Is there anything we can do to help with your PR?

[pdeziel]

@pdeziel bump Is there anything we can do to help with your PR?

Hi @lwgray, I think I have most of the functionality completed, although I still need to clean up the code comments, add documentation, etc.

There is one visual bug I was working on tracking down which I might need help with; the color mapping on the visualization doesn't seem correct (e.g., 0.29 in the below chart should be "bluer" than 0.16). This might have to do with pcolormesh expecting a value in the range [0,1] but the correlation value range is from [-1,1]. I would also ideally like to have the red and blue values swapped in the default color map (so something like BuYlRd) but that color map doesn't currently exist in YB.

[lwgray]

@pdeziel Awesome work... Maybe you can internal convert it to be between 0 and 1 while still display -1 to 1. Maybe @bbengfort or @rebeccabilbro can chime in on this. I guess in the meantime you can work on the docs. Also, make sure you update your branch.

[bbengfort]

@pdeziel @lwgray I'm wondering if the colors bug is a mismatch of the labels and the colors; e.g. a matrix got inverted somewhere incorrectly. If you look at (Tatsumi Kimishima, Nintendo) = .29, c=yellow; and (woman, play) = 0.04 c=blue, these cells are inverse of each other on the matrix but seem to have the other cell's color.

I'd look at the wc_display flip on L174 as a start, then check the top left and top right positions in X and Y.

[pdeziel]

@lwgray @bbengfort I've fixed the color mapping issue and added the documentation, so I think is ready for review again. There seems to be one CI error, but it's in a different module so I haven't figured out how it's related to my changes yet.

One thing I think would be useful to add in the future is a parameter where the user would only have to supply one word, and the visualizer finds the top correlated words in the corpus.