Release 3.2.3

[tovbinm]

Changes from 3.2.0 to 3.2.1:

• Upgraded to Spark 2.2.1

Changes from 3.2.1 to 3.2.2:

• Numeric bucketizers fixes: including metadata and integration with sanity checker.
• Introduced split inclusion parameter for numerical bucketizers controlling should the splits be left or right inclusive (numeric.bucketize(splitInclusion) and numeric.autoBucketize(splitInclusion)
• Introduced an option for numerical bucketizers to allow tracking invalid values such as NaN, -Inf, Inf or values that fall outside the buckets. (numeric.bucketize(trackInvalid) and numeric.autoBucketize(trackInvalid)).
• Finalized the unification of vectorization for ALL the OP types, including Text, TextArea, Base64, Phone, URL and Geolocation.
• Fixed track null behavior of vectorization for MulitPicklistMap to match MultiPicklist
• OP cli --auto properly identifies the input csv schema as expected
• OP cli generated code is now prettier
• Minor bug fixes in vectorizers metadata and ctor args for sanity checker model

Changes from 3.2.2 to 3.2.3:

• Added null tracking for map vectorizers
• Added null tracking for hashed text features
• Sanity Checker: fixed issue in feature removal for sibling features when one has a correlation of NaN with the label
• Fixes for Decision Tree bucketizer vector metadata to allow it working with Sanity Checker
• Correlation based record level insights
• All model classes are now made public with private[op] ctors
• Added inclusion indicators in bucket labels
• Workflow now creates holdout during fitting after raw data creation and applies eval
• Simplified OP CLI generated template
• Added VersionInfo - which allows access to project version and git info at runtime and include it in AppMetrics
• Make most OpWorkflowRunner ctor params optional and deprecate old ctors
• OpWorkflowRunner.run now requires spark StreamingContext to be present
• Implemented a new run type allowing streaming score (--run-type=streamingScore)

[Jauntbox]

LGTM!