[
https://issues.apache.org/jira/browse/SPARK-55312?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-55312:
-----------------------------------
Labels: pull-request-available (was: )
> Always infer Spark types in createDataFrame when schema is not provided
> -----------------------------------------------------------------------
>
> Key: SPARK-55312
> URL: https://issues.apache.org/jira/browse/SPARK-55312
> Project: Spark
> Issue Type: Improvement
> Components: PySpark
> Affects Versions: 4.2.0
> Reporter: Yicong Huang
> Priority: Major
> Labels: pull-request-available
>
> Currently, when {{createDataFrame}} is called without a schema, the
> {{spark_types}} list contains {{None}} for non-timestamp columns:
> {code:python}
> # pyspark/sql/pandas/conversion.py
> spark_types = [
> TimestampType()
> if is_datetime64_dtype(t) or isinstance(t, pd.DatetimeTZDtype)
> else None # Non-timestamp columns get None
> for t in pdf.dtypes
> ]
> {code}
> This {{None}} is passed to the serializer, which then lets PyArrow infer the
> type. This creates inconsistency because the Spark type is not explicitly
> defined.
> The fix is to always infer Spark types from Arrow types when no schema is
> provided:
> {code:python}
> arrow_schema = pa.Schema.from_pandas(pdf, preserve_index=False)
> spark_types = [
> TimestampType()
> if is_datetime64_dtype(t) or isinstance(t, pd.DatetimeTZDtype)
> else from_arrow_type(field.type, prefer_timestamp_ntz)
> for t, field in zip(pdf.dtypes, arrow_schema)
> ]
> {code}
> This ensures {{spark_type}} is always defined, which:
> 1. Makes the code more consistent
> 2. Enables downstream code to always have a valid Spark type
> 3. Prepares for making {{spark_type}} a required parameter in serializers
> Files to change:
> - {{pyspark/sql/pandas/conversion.py}}
> - {{pyspark/sql/connect/session.py}}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
