[
https://issues.apache.org/jira/browse/ARROW-8004?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17055776#comment-17055776
]
Joris Van den Bossche commented on ARROW-8004:
----------------------------------------------
For a more limited use case than general objects, i.e. array-like objects, we
could also think about checking for {{__array__}} and then convert to ListArray.
For example, now we can infer a list of lists or a list of numpy arrays, but
will error on a list of pandas Series
> [Python] Define API for user-defined conversions of array cell values in
> pyarrow.array
> --------------------------------------------------------------------------------------
>
> Key: ARROW-8004
> URL: https://issues.apache.org/jira/browse/ARROW-8004
> Project: Apache Arrow
> Issue Type: Improvement
> Components: Python
> Reporter: Wes McKinney
> Priority: Major
>
> Consider the statement
> {code}
> pyarrow.array([v0, v1, v2, v3])
> {code}
> or correspondingly
> {code}
> pyarrow.array(pd.Series([v0, v1, v2, v3], dtype=object))
> {code}
> where {{v0, ..., v4}} are instances of types with no built-in
> conversion-to-Arrow support in pyarrow. An API could be provided to allow
> user-defined unboxing to a data type that the library _does_ understand (like
> a NumPy array). One complexity is that if the unboxing is costly, we may need
> to "keep around" the unboxed value when doing multiple passes over the data
> (e.g. initially for type inference and then for conversion)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
