[ML] Allow scientific notation for formatting very small actual and typical values

[richcollier]

Kibana version:
6.6.2

Notice the typical column:

Here's the associated anomaly record:

    "hits" : [
      {
        "_index" : ".ml-anomalies-shared",
        "_type" : "doc",
        "_id" : "io_monitoring_record_1544950800000_900_0_906066770_33",
        "_score" : 0.0,
        "_source" : {
          "job_id" : "io_monitoring",
          "result_type" : "record",
          "probability" : 2.530814467762354E-230,
          "multi_bucket_impact" : -5.0,
          "record_score" : 98.3941490916286,
          "initial_record_score" : 98.3941490916286,
          "bucket_span" : 900,
          "detector_index" : 0,
          "is_interim" : false,
          "timestamp" : 1544950800000,
          "partition_field_name" : "source.keyword",
          "partition_field_value" : "xxxxxxx.xxxxxxx.com",
          "function" : "mean",
          "function_description" : "mean",
          "typical" : [
            1.0758673682845221E-12
          ],
          "actual" : [
            0.002499999944120646
          ],
          "field_name" : "value",
          "influencers" : [
            {
              "influencer_field_name" : "source.keyword",
              "influencer_field_values" : [
                "xxxxxxx.xxxxxxx.com"
              ]
            }
          ],
          "source.keyword" : [
            "xxxxxxx.xxxxxxx.com"
          ]
        }
      }
    ]

[elasticmachine]

Pinging @elastic/ml-ui

[peteharverson]

The actual and typical values are formatted according to the Kibana format that has been configured for the field from the corresponding index pattern. If no format has been specified for the particular index pattern field (on the index patterns page under the Kibana Management tab), it will use the number format, configured under Management - Advanced Settings, which defaults to 0,0.[000], which is what is happening in the screenshot above.

The number format uses the numeral.js format, but Kibana uses a fork of numeraljs, which unfortunately does not yet support the exponential format which is now provided by numeral.js. An enhancement request is open to add support for scientific format - see #5811.

So until exponential format is added to the fork of numeral.js, I don't believe there is much we can offer here, as I think using the Kibana-wide field formats offers more benefits than ML providing its own custom formatting. If you wanted to drop the comma separators, you could configure a format of 0 for that field in the index pattern to give e.g.