[jira] [Commented] (ARROW-1660) [Python] pandas field values are messed up across rows
[ https://issues.apache.org/jira/browse/ARROW-1660?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16219432#comment-16219432 ] MIkhail Osckin commented on ARROW-1660: --- I definitely tested it with the latest pyarrow version at the moment. I had the same intuition that this issue might be related to splicing, because my initial dataset was ordered by id field and top of the dataset (after to_pandas) was something like this 10012, 10015, 10034, and the row with id like 10018 had values from 100034 and only part of them at least in one column (and if i remember well 10018 was the exact third id by ascendence. > [Python] pandas field values are messed up across rows > -- > > Key: ARROW-1660 > URL: https://issues.apache.org/jira/browse/ARROW-1660 > Project: Apache Arrow > Issue Type: Bug > Components: Python >Affects Versions: 0.7.1 > Environment: 4.4.0-72-generic #93-Ubuntu SMP x86_64, python3 >Reporter: MIkhail Osckin >Assignee: Wes McKinney > > I have the following scala case class to store sparse matrix data to read it > later using python > {code:java} > case class CooVector( > id: Int, > row_ids: Seq[Int], > rowsIdx: Seq[Int], > colIdx: Seq[Int], > data: Seq[Double]) > {code} > I save the dataset of this type to multiple parquet files using spark and > then read it using pyarrow.parquet and convert the result to pandas dataset. > The problem i have is that some values end up in wrong rows, for example, > row_ids might end up in wrong cooVector row. I have no idea what the reason > is but might be it is related to the fact that the fields are of variable > sizes. And everything is correct if i read it using spark. Also i checked > to_pydict method and the result is correct, so seems like the problem > somewhere in to_pandas method. -- This message was sent by Atlassian JIRA (v6.4.14#64029)
[jira] [Commented] (ARROW-1660) [Python] pandas field values are messed up across rows
[ https://issues.apache.org/jira/browse/ARROW-1660?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16219171#comment-16219171 ] Wes McKinney commented on ARROW-1660: - Is it possible you were using pyarrow < 0.7.0? There was a bug ARROW-1357 that was fixed that would cause the issue you were seeing. I'm a bit at a loss since the relevant test case is https://github.com/apache/arrow/blob/master/python/pyarrow/tests/test_convert_pandas.py#L600. I will move off the 0.8.0 milestone, but leave the issue open in case you can find a repro > [Python] pandas field values are messed up across rows > -- > > Key: ARROW-1660 > URL: https://issues.apache.org/jira/browse/ARROW-1660 > Project: Apache Arrow > Issue Type: Bug > Components: Python >Affects Versions: 0.7.1 > Environment: 4.4.0-72-generic #93-Ubuntu SMP x86_64, python3 >Reporter: MIkhail Osckin >Assignee: Wes McKinney > > I have the following scala case class to store sparse matrix data to read it > later using python > {code:java} > case class CooVector( > id: Int, > row_ids: Seq[Int], > rowsIdx: Seq[Int], > colIdx: Seq[Int], > data: Seq[Double]) > {code} > I save the dataset of this type to multiple parquet files using spark and > then read it using pyarrow.parquet and convert the result to pandas dataset. > The problem i have is that some values end up in wrong rows, for example, > row_ids might end up in wrong cooVector row. I have no idea what the reason > is but might be it is related to the fact that the fields are of variable > sizes. And everything is correct if i read it using spark. Also i checked > to_pydict method and the result is correct, so seems like the problem > somewhere in to_pandas method. -- This message was sent by Atlassian JIRA (v6.4.14#64029)