arrow-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Wes McKinney (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ARROW-86) Implement zero-copy Arrow-to-Pandas conversion
Date Mon, 28 Mar 2016 16:20:25 GMT

    [ https://issues.apache.org/jira/browse/ARROW-86?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15214400#comment-15214400
] 

Wes McKinney commented on ARROW-86:
-----------------------------------

Note that because numpy.ndarray does not have any notion of null / NA, that if {{null_count
> 0}} then a copy is necessary. For numeric or boolean data without any nulls, zero-copy
would be great.

> Implement zero-copy Arrow-to-Pandas conversion
> ----------------------------------------------
>
>                 Key: ARROW-86
>                 URL: https://issues.apache.org/jira/browse/ARROW-86
>             Project: Apache Arrow
>          Issue Type: Improvement
>          Components: Python
>            Reporter: Uwe L. Korn
>
> For this to work, we need:
>  * a check if the Arrow Array could be represented as Pandas Series using the same memory
buffer
>  * a {{numpy.ndarray}} subclass that keeps a reference to the Arrow Array
> With the NumPy C-API it is already possible to create a {{numpy.ndarray}} instance that
uses the Arrow Array memory buffer internally but there is no way to keep track if the {{numpy.ndarray}}
still lives or the underlying memory can be free'd.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message