[ML] Data Frame Analytics: ROC Curve Chart

[walterra]

Meta Issue for ROC Curve Chart

7.12

• Chart for Classification Results Pages [ML] Data Frame Analytics: ROC Curve Chart #89991
• Functional Tests

7.13

• Fix ROC Curve Chart for binary classification. [ML] Data Frame Analytics: Fix ROC Curve Chart for binary classification. #94791

Discuss

• It's worth discussing the behavior of filtering the data used to create the chart. At the moment, by default, no filter gets applied to the query. The results page UI has a query bar and the training/testing preset buttons. Custom queries and these preset also get added to the query used to fetch the ROC/AUC data. This gives us the most flexibility, but is it the best user experience? Does a data scientist expect to have these options or would they only query e.g. with the testing filter anyway?

Todos

• Chart for Outlier Detection Results Pages — requires additional UI to select a boolean field with expected outlier training data.
• Investigate for using a responsive layout for the Model evaluation panel, putting the ROC curve to the right of the confusion matrix in a second column if there is space
• Align colors with feature importance chart
• If a search filter results in 0 docs, avoid calling the ROC API endpoint and display a blue info callout instead

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[tveasey]

I think user defined queries would be useful. The key use case is applying a semantic filter (for example a particular DGA family) and observing the ROC curve and its AUC for just those test data.