Running training when dataset only has matches/non matches or limited samples throws errors. We should instead inform the user about this so they can add training samples.

[sonalgoyal]

Reported by Luke from Databricks

zingg_Dec21_0823_log4j-active (1).txt zingg_Dec21_0823_sdtderr.txt

[navinrathore]

When there is no training data available, NullPointerException is thrown.

[navinrathore]

Other problematic scenarios: (refer to attached log file)

1. When only negative or positive training data are available.
2. When less number of training data are available. Error 1: java.lang.IllegalArgumentException: requirement failed: rawPredictionCol vectors must have length=2, but got 1 Error 2: java.lang.IllegalArgumentException: requirement failed: Nothing has been added to this summarizer

An appropriate error message should be added to ask user to add more training data

[sonalgoyal]

let's fix all

[sonalgoyal]

@gnanaprakash-ravi please verify this

[gnanaprakash-ravi]

Hi,

1. when trainingData is null

2. when neg is null and pos is less than 5 result:

3. When pos is equal to 5 and neg is null result:

4. when pos and neg are exactly equal to 5 in the train phase (Needs to be analyzed intensively) This behavior is occurring on the new model and new zinggdir I suspect this error might be related to Apache Spark library but this was intercepted by zinggbusinessexception: (after the code change)

[sonalgoyal]

1,2,3 are working as expected. 4 is giving an exception with the error around less data. No fix needed.