Release 2.1.0

What Changed?

• Added support for SKlearn model deployment using jep
• Fixed PySpliceContext functions to use underlying Scala Implementations for more efficient execution
• Database transaction ID is now a LONG instead of an INT
• Code cleanup

This release is in tandem with ml-workflow

HOT FIX for DB Transaction ID

INT -> LONG

Robust support for model libs

Support for saving and loading sklearn, spark, h2o and keras models
This release is in tandem with ml-workflow

H2O In Database Model Deployment

Support for deploying (supported) H2O models directly into the database (like spark models)

H2O added support

H2O Model saving and loading, as well as context manager support for MLFlow

Update to plotROC and start_run

PlotROC had chart axes flipped

mlflow.start_run doesn't set the run_name if you pass it in. Fix for that issue

New MLManager API

BREAKING CHANGES: MLManager has been completely redesigned and the MLManager class no longer exists.

The new implementation gives users full functionality to the vanilla mlflow api with mlflow.FUNCTION calls, as well as Splice's custom functions through the same module.

Accessing the mlflow module is simply from splicemachine.mlflow_support import *
And to handle artifacts, simply call mlflow.register_splice_context(splice)

MLFlow upgrade:
@abaveja313 and @Ben-Epstein

MLFlow Upgrades

MLFlow 1.6.0 Upgrade
Ability to log zip files as artifacts to mlflow
Ability to download artifacts from the MLFlow UI (including zip files)
analyzeTable improvements (@jpanko1)

Spark Statistical tools added + MLManager schema addressed

Many statistical tools like Oversampler and OversamplerCrossValidator among others - @sharmanirek
Bug fix in evaluator class. Now references user assigned label and prediction columns - @sharmanirek
Updated default mlmanager schema to MLMANAGER from SPLICE

In DB Deployment partial release

This release is to handle the partial merge of the db deployment code