[jira] [Commented] (ARROW-3947) [Python] query distinct values of a given partition from a ParquetDataset

["Joris Van den Bossche (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-3947?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17608770#comment-17608770 ] 

Joris Van den Bossche commented on ARROW-3947:
----------------------------------------------

Note that with the latest version of the dataset API, there is a {{partitioning}} attribute (instead of {{pieces}}), which can also be used to retrieve this information:

{code:python}
import pyarrow as pa
import pyarrow.parquet as pq

table = pa.table({'a': range(9), 'part': ['a', 'b', 'c']*3})
pq.write_to_dataset(table, "test_partitioned", partition_cols=["part"])

dataset = pq.ParquetDataset("test_partitioned", use_legacy_dataset=False)

>>> dataset.partitioning
<pyarrow._dataset.HivePartitioning at 0x7fe760a1f580>

>>> dataset.partitioning.dictionaries
[<pyarrow.lib.StringArray object at 0x7fe740b6fdc0>
 [
   "a",
   "b",
   "c"
 ]]
{code}

Personally, I think this is sufficient API around this, but we should certainly do a better job documenting this.

> [Python] query distinct values of a given partition from a ParquetDataset
> -------------------------------------------------------------------------
>
>                 Key: ARROW-3947
>                 URL: https://issues.apache.org/jira/browse/ARROW-3947
>             Project: Apache Arrow
>          Issue Type: New Feature
>          Components: Python
>    Affects Versions: 0.10.0
>         Environment: MacOSX, Python 3.6, 
>            Reporter: Ji Xu
>            Priority: Minor
>              Labels: dataset-parquet-read, parquet
>
> Right now the values of a given partition from a `ParquetDataset` is buried inside `ParquetDataset.pieces`, a bit inconvenient for the user to dig out this information. A helper function/method to perform this task in `ParquetDataset` class would be very helpful for the users.
> A pure personal opinion on the name of this method: `ParquetDataset.select_distinct()` with partition_name as the positional arg, to resemble SQL `SELECT DISTINCT column FROM table`.
> I'm not sure how to contribute here on Jira, so I created this [GitHub Gist |https://gist.github.com/xujiboy/c3fcc47f720ed9adf2260c5d0ba8aed2]as an possible solution for this problem.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)