Datarobot-batch-scoring

======================

* New semantic routes versioning support

* New prediction response schema support

* **Dropped support of DataRobot Prediction API < 3.0 version.**

========================

* Independent prediction service support for scoring

======================

* switched to supervisor + workers architecture, improving handling of errors and
subprocess lifecycle control.

* Source code split into more mostly isolated modules.

=======================

* added 3rd parallel process which handles post-processing and writing of responses.
This should greatly improve performance.

* add ability to compress data in transit

========================
* --output_delimiter flag to set delimiter for output CSV. "tab" can be used
for tab-delimited output

* --skip_row_id flag to skip row_id column in output

* fixed hang of batch-scoring script on CSV parse errors

* added summary of run at the end of script output with full list of errors,
warnings and total stats.

* fixed error when trying to report multiline CSV error in fast mode

* Run all tests against Windows

=======================
* --pred_name parameter is documented. Potentially backward incompatible change:
Previously, 1.0 class was used as positive result for binary predictions,
now last class in lexical order is used

* Fixed memory leak and performance problem caused by unrestricted batch-generator

* internal check and error avoidance logic for requests that are too large

* docker and docker-compose files for dockerized run of tests and script

* auto sampler target batch size increased to 2.5M