v0.6.6

• Align fixed windows first and last buckets time to the window size

v0.6.5

• Allow pyarrow 3.x

v0.6.4

• set max_value to new aggregations
• ML-691: start_filter should be >= (and not >)
• ML-717: Use logger.debug instead of print.

v0.6.3

• Flatten/copy bug issue fix
• Fixed window rework
• ML-707: bug fix in multiple integer key columns
• ML-701: Fix ordinary timestamp type fields in ingestion service.
• ML-635: unexpected results aggregating with sliding window
• ML-673: removing adding of "igz_part" to partitions

v0.6.2

• Support typed indices in writers.
• None value should remain None in CsvSource

v0.6.1

• ML-602: Typed columns for explicitly setting parquet schema. (#231)
• ML-613: Filter out value for empty key (#230)
• AsyncEmitSource – async def run -> def run (#234)
• ML-667: Hash calculation should be modulo instead of division (#235)

v0.6.0

• First/last aggregations
• bug fix in reading non existing column from v3io when keys are list…

v0.5.4

• change default number of buckets in fixed windows

v0.5.3

• update v3io-py version
• ML-625: csv - discovering type if None is in first row

v0.5.2

• ML-612: supporting separate aggregation steps
• Allow older numpy versions.