TEST: using get_dataframes_and_queues instead of get_queues in onedal4py testing
#1909
Conversation
added docstring for get_queues added docstring for get_dataframes_and_queues
update test_incremental_covariance.py
|
/intelci: run |
|
/intelci: run |
samir-nasibli
added
the
testing
|
/intelci: run |
|
/intelci: run |
|
/intelci: run |
|
/intelci: run |
There was a problem hiding this comment.
Since np_sycl isn't a dataframe type unlike array_api, pandas, numpy, dpctl, or dpnp, I'd ask that you write comments in the code in the instances where np_sycl is used to give future developers the reasoning why its there/ flag it for special observation. I'd also add some comments into get_dataframes_and_queues. I had been thinking why we just don't extend numpy to include queues. Did you run that and see how much it increases CI runtimes?
Yeah, make sense to share statistics of the run. Also definitely docs needed to be updated. Thank you! |
|
This will need to be rebased for 2025.0 |
|
Some updates are required. Moving back to the draft. |
|
@samir-nasibli what needs to be done to get out of Draft state? |
| if "np_sycl" in dataframe_filter_: | ||
| # sycl queue param is not None. | ||
| # Designed for interfaces that utilize NumPy inputs with a DPCTL queue, | ||
| # enabling offloading to specific SYCL devices. | ||
| dataframes_and_queues.extend(get_df_and_q("numpy_and_queue")) |
There was a problem hiding this comment.
-
what do you think about the naming and this approach specifically? Does it make sense to leave only
numpy_and_queueinsteadnp_sycl? -
Maybe it is better to use param:
numpy_with_queue=True
I mean:
get_dataframes_and_queues("numpy", numpy_and_queue=True))
instead of:
get_dataframes_and_queues("numpy,np_sycl"))
There was a problem hiding this comment.
the second option could be generalized for all dataframes that doesn't support sycl_queue, but when sycl_queue is provided
|
/intelci: run |
Description
using
get_dataframes_and_queuesinstead ofget_queuesin onedal4py testing.Enabled only for covariance, incremental covarience, incremental pca (just for verifying the primitive work)
Justification
onedal4py API and SPMD APIs have ability to work with
numpyandsycl_queueprovided. Currently we are testing them with numpy with None queue or with dpctl/dpnp with None queue param. This PR updates get_dataframes_and_queues testing primitive, enhancement includes casesnumpy+sycl_queue.TODO
Separately enable it for all cases where
get_queuesused + SPMD iface (#1777 )Should be