Github user BryanCutler commented on a diff in the pull request: https://github.com/apache/spark/pull/19646#discussion_r148709143 --- Diff: python/pyspark/sql/session.py --- @@ -416,6 +417,50 @@ def _createFromLocal(self, data, schema): data = [schema.toInternal(row) for row in data] return self._sc.parallelize(data), schema + def _getNumpyRecordDtypes(self, rec): + """ + Used when converting a pandas.DataFrame to Spark using to_records(), this will correct + the dtypes of records so they can be properly loaded into Spark. + :param rec: a numpy record to check dtypes + :return corrected dtypes for a numpy.record or None if no correction needed + """ + import numpy as np + cur_dtypes = rec.dtype + col_names = cur_dtypes.names + record_type_list = [] + has_rec_fix = False + for i in xrange(len(cur_dtypes)): + curr_type = cur_dtypes[i] + # If type is a datetime64 timestamp, convert to microseconds + # NOTE: if dtype is M8[ns] then np.record.tolist() will output values as longs, + # this conversion will lead to an output of py datetime objects, see SPARK-22417 + if curr_type == np.dtype('M8[ns]'): --- End diff -- There shouldn't be any difference for the most part. I only used `M8` here because when debugging these types, that is what was being output for the record types by `numpy.record.dtype`. Would you prefer `datetime64` if that works?
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org