Estimator Score Visualizers

[bbengfort]

No description provided.

[rebeccabilbro]

Implement a base class that will follow an estimator in a pipeline (keep in mind that there could be more than one of these things stacked up along each other). This is not a transformer. It needs to extend BaseEstimator, VisualizerMixin. We also need to create a ScoringMixin. ScoringMixin has scoring function that saves the state of the scoring to the class so that we can draw it. This will have:

• fit
• predict
• score
• draw

Need to think how we'll access the model from ScoreVisualizer. Best option is to have it be instantiated with a model form - this makes sense because we would call fit down into the estimator. The trick will be that the visual pipeline needs to know not to run fit twice.

[rebeccabilbro]

@bbengfort ok, preliminary implementation here: https://github.com/DistrictDataLabs/yellowbrick/blob/develop/yellowbrick/base.py#L104

[rebeccabilbro]

ClassifierScoreVisualizers:

• ClassificationReport
• ROCAUC

RegressionScoreVisualizers:

• PredictionError
• ResidualsPlot

[bbengfort]

@rebeccabilbro ok I took a look at these, and things are looking great! There are a few things that we should resolve/discuss and this issue will be done!

Also - I really like how you've separated score and draw and poof - it was the basis for everything I did in the feature visualizers!

• decide on self.classes or self.classes_ - I guess I was thinking we'd act a bit more like Scikit-Learn; I'm open to other opinions.
• In order to make the isestimator check happen (from the super class), we'll have to call super in __init__
• We should elevate the model wrapping logic to ScoreVisualizer as in Refactor all Visualizers to move fig/ax hooks to base class #62
• What is the difference between ScoreVisualizer and ModelVisualizer?

[rebeccabilbro]

@bbengfort have addressed the first two checkboxes in commits d57102c and 76d2d1f

Will tackle the third in the next sprint.

As for the 4th, I've clarified the difference in base.py here. Essentially the main difference is that ScoreVisualizer takes a fitted model and ModelVisualizer takes a model form.

[rebeccabilbro]

Marking as complete, with the understanding that issue #62 will resolve the fig/ax hook question that came up in the design of this class.