[ML] Meta - Classification UI

[alvarezmelissa87]

• Classification 
  • Create
    • [ML] DF Analytics: create classification jobs via the UI #51619 *
    • classification job type help text (must have)
    • add classification job type to dropdown in create flyout (must have) *
    • User can create classification job using basic form with the fields(must have):
      • User can enter job ID (must have) *
      • User can enter source index (must have) *
      • User can enter destination index (must have) *
      • User can enter dependent_variable (must have) *
        • Fetch options using field_caps api - grab all allowed type fields as options *
      • User can enter training percent (must have) *
      • User can enter job description (must have)
        • [ML] DF Analytics: add description field to job creation and display in jobs list #52217
      • User can select to included/excluded fields in basic form - use explain API (should have)
        • [ML] DF Analytics job creation: Add 'excludes' input field to form #53856
      • [ ] disable the Create button in the advanced editor if there is a syntax error in the JSON (removed as code editor already provides validation)
  • Evaluate
    • Once the classification job has completed, the results of the job should be evaluated
      • [ML] DF Analytics: create classification jobs results view #52584
    • The evaluation step should be run automatically (should have)
      • [ML] DF Analytics: create classification jobs results view #52584
    • Ensure results view panels are responsive
    • Fix positioning of labels on confusion matrix
      • [ML] DF Analytics Results: adds link to docs #54189
    • Add link to docs in top right corner of panel
      • [ML] DF Analytics Results: adds link to docs #54189
    • update to account for changes to the _evaluate endpoint introduced in Emit predicted category using an appropriate JSON type. ml-cpp#877 in which the .keyword suffix will no longer be needed for boolean or integer dependent variable fields
      • [ML] DF Analytics Classification: ensure confusion matrix can be fetched #53629
  • Explore
    • The user can view the evaluate results (must have)
      • [ML] DF Analytics: create classification jobs results view #52584

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[alvarezmelissa87]

Would you be up for taking a look at the checklist I've got so far? Feel free to point out anything I've missed or have gotten wrong. I've got a couple of clarification questions below, as well. Thank you!
cc @peteharverson, @sophiec20

1. Do we want num_top_classes available in the form or just in the advanced editor?

2. It looks like there are 2 options when evaluating classification jobs (https://www.elastic.co/guide/en/elasticsearch/reference/7.5/evaluate-dfanalytics.html#_binary_soft_classification). Is which one to use dependent on the value in num_top_classes?
   E.g. If 2 (the default) then it's binary_soft_classification, otherwise use classification? Or is the type of evaluation something the user will have to specify?

[peteharverson]

My thoughts on #51310 (comment) are

1. num_top_classes should not be available in the form. If only binary classification is currently done by the back-end currently, I wonder if it even needs to be specifically added in the advanced editor?

2. The evaluation API we should be using is the classification one documented at https://www.elastic.co/guide/en/elasticsearch/reference/7.5/evaluate-dfanalytics.html#_classification_2. We need to be displaying the confusion matrix in some form in the results view, along with initially some form of tabular view where the user can compare the predicted class to the actual class.

[sophiec20]

The num_top_classes back-end default is 2. For binary classification, there is little point in making this any more visible to the end user as we do not expect them to need to change this value.