1239 visualize decision trees

[ParthivNaresh]

Fixes #1239

---

The output of visualize_decision_tree() will be a graphviz.files.Source object.

For example visualize_decision_tree(clf=regression_estimator, filled=True, max_depth=2).view()

The output of clean_format_tree will be an OrderedDict.

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #1511 (47adf09) into main (3de3b12) will increase coverage by 0.1%.
The diff coverage is 100.0%.

@@            Coverage Diff            @@
##             main    #1511     +/-   ##
=========================================
+ Coverage   100.0%   100.0%   +0.1%     
=========================================
  Files         232      232             
  Lines       16430    16639    +209     
=========================================
+ Hits        16422    16631    +209     
  Misses          8        8             

Impacted Files	Coverage Δ
evalml/model_understanding/graphs.py	99.7% <100.0%> (+0.1%)	⬆️
...s/estimators/regressors/decision_tree_regressor.py	100.0% <100.0%> (ø)
evalml/tests/conftest.py	100.0% <100.0%> (ø)
...lml/tests/model_understanding_tests/test_graphs.py	100.0% <100.0%> (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 3de3b12...47adf09. Read the comment docs.

[dsherry]

@ParthivNaresh I'm excited to review this! Could you please include an example of the output in the PR description? Will make it quick for reviewers to understand what's up.

[freddyaboulton]

@ParthivNaresh Looks great! And tests look solid. I left some comments but I think the only thing holding merge is the discussion about how to display the column names in the clean_format_tree

[freddyaboulton]

I think we should save the actual feature names in the output. These feature names range from 0 to n_col-1 because we convert from pandas to sklearn right before fitting. The problem is that the feature names are not saved in the tree object.

I think we have a couple of options:

1. Add an option to this function for passing in the feature names
2. Change the input type from a tree estimator to a pipeline with a tree estimator. This would allow us to use input_feature_names[estimator.name]
3. File an issue for saving the feature names to the tree estimator and leaving this function as-is for now.

I think I prefer 2 but what do you think? I believe our model understanding methods take in a pipeline instead of an estimator so it'd be more consistent with what we already have. I guess we can also do both 1 and 2, where we add the feature names as parameter to this function and then add another function that accepts a pipeline that calls this function.

[bchen1116]

Looks great! I added a few suggestions on additional tests, but nothing blocking.

[dsherry]

@ParthivNaresh this is awesome!

I left a few suggestions and questions. Ones I'd like us to address before merge:

• Method naming
• Use data method in graph method
• Split graph unit tests into checking the returned graph content vs checking the filepath image saving

I also left a note about #1535 , that could be cool to look at next!

[angela97lin]

Looks really good! Just added some tiny nit-picky comments about breaking down tests more and docstrings.