[ARROW-5655] [Python] Table.from_pydict/from_arrays not using types in specified schema correctly

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.16.0
• Component/s: Python
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/22090
• Assignee: Krisztian Szucs
• Reporter: Joris Van den Bossche
• Created: 19/Jun/19 18:34
• Updated: 11/Jan/23 07:41
• Resolved: 08/Oct/19 13:33

Description

Example with from_pydict (from https://github.com/apache/arrow/pull/4601#issuecomment-503676534):

In [15]: table = pa.Table.from_pydict(
    ...:     {'a': [1, 2, 3], 'b': [3, 4, 5]},
    ...:     schema=pa.schema([('a', pa.int64()), ('c', pa.int32())]))

In [16]: table
Out[16]: 
pyarrow.Table
a: int64
c: int32

In [17]: table.to_pandas()
Out[17]: 
   a  c
0  1  3
1  2  0
2  3  4

Note that the specified schema has 1) different column names and 2) has a non-default type (int32 vs int64) which leads to corrupted values.

This is partly due to Table.from_pydict not using the type information in the schema to convert the dictionary items to pyarrow arrays. But then it is also Table.from_arrays that is not correctly casting the arrays to another dtype if the schema specifies as such.

Additional question for Table.pydict is whether it actually should override the 'b' key from the dictionary as column 'c' as defined in the schema (this behaviour depends on the order of the dictionary, which is not guaranteed below python 3.6).

Issue Links

• links to: GitHub Pull Request #5567