ovalhub commented on issue #10665:
URL: https://github.com/apache/lucene/issues/10665#issuecomment-1297463120
On Mon, 31 Oct 2022, Benjamin Trent wrote:
> @ovalhub `numpy` collections are already native. To use them, I have to
> pull them into python collections and then cast them to be native again.
>
> Example: ``` X = X.tolist() doc.add(KnnVectorField("knn",
> JArray('float')(X), fieldType)) ``` The `.toList()` is weird to me as `X`
> is already a natively backed float array, just a numpy object.
>
> Admittedly, I may be misunderstanding how the communication between Python
> -> `native` -> Java is done. But it SEEMS that calling `tolist` for numpy
> unnecessarily pulls an already native collection into a generic python
> list.
When you make a JArray with "JArray('float')(X), fieldType))" the
JArray<float> code in JCC's JArray.h will iterate 'X' "as a sequence",
calling PySequence_GetItem(sequence, i) and fill a Java float array with the
values returned.
If native numpy collections are exposed to Python as sequences (ie, ordered
things that can be accessed with int indexes), then you should not have to
call .tolist() on it before using it with PyLucene/JCC.
It could also be that numpy's .tolist() just creates a wrapper object around
the native array to expose it as a Python sequence ? (I don't know numpy).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
