Include confusion matrix threshold details in Performance and Prediction Section

[exalate-issue-sync]

+Details:+
Let's update the [Prediction and Performance section|http://docs.h2o.ai/h2o/latest-stable/h2o-docs/performance-and-prediction.html] to include details on how the confusion matrix threshold is applied. From the [code|https://github.com/h2oai/h2o-3/blob/2a6fdc715e8c4317df7320eaddf8bae2524889fa/h2o-genmodel/src/main/java/hex/genmodel/GenModel.java#L329] you can see that we predict the positive class (1, True, or categorical label that is second in the lexicographical ordering of the labels) if the predicted probability is *greater than or equal * to the threshold, which produces the best F1 score.

Documentation Update:
In the [confusion matrix section|http://docs.h2o.ai/h2o/latest-stable/h2o-docs/performance-and-prediction.html#confusion-matrix]

Below the paragraph and image in that section we should add:

Binary Classification
All predicted probabilities greater than or equal to the F1 Max threshold are labeled with the positive class (e.g., 1, True, or the second label in lexicographical order). The F1 Max threshold is selected to maximize the F1 score calculated from confusion matrix values (true positives, true negatives, false positives, and false negatives).

Multiclass Classification
Prediction class labels are based on the class with the highest predicted probability.

[exalate-issue-sync]

Angela Bartz commented: Pull request submitted into rel-yule.

[exalate-issue-sync]

Angela Bartz commented: Pull request merged into rel-yule.

[h2o-ops]

JIRA Issue Migration Info

Jira Issue: PUBDEV-7407
Assignee: Angela Bartz
Reporter: Lauren DiPerna
State: Resolved
Fix Version: 3.30.0.1
Attachments: N/A
Development PRs: Available

Linked PRs from JIRA

#4464