Make KernelRidge inherit from UniversalBase #6327
Conversation
Support for 0cc
viclafargue
left a comment
There was a problem hiding this comment.
LGTM for the most part, but I would add a test in the test_device_selection.py file to ensure that training and inference can be done on both CPU or GPU and the conversion between the two is seamless.
|
|
||
| @generate_docstring() | ||
| @enable_device_interop | ||
| def fit(self, X, y, sample_weight=None, |
There was a problem hiding this comment.
Also, small detail, but it is always nice to use the self.n_features_in_ attribute instead of local variable n_rows when ingesting data. This gives one more thing to replicate sklearn's behavior.
There was a problem hiding this comment.
You mean we should set self.n_features_in_ instead of using a local variable? Can do that. One question, n_features_in_ is the number of features, so shouldn't it be set to n_cols instead?
There was a problem hiding this comment.
You're right meant the number of columns.
|
/merge |
Previously this wasn't handled correctly, but was _sometimes_ masked by `GridSearchCV` in the old test fixing things behind the scenes. This fixes the bug, adds a test that doesn't require the accelerator, and changes the accelerator test to be both simpler (matching the smoketests in the rest of the file) and much faster (the `GridSearchCV` in the test made the test quite slow on my machine). Fixup to the fix applied in #6327. Authors: - Jim Crist-Harif (https://github.com/jcrist) Approvers: - Dante Gama Dessavre (https://github.com/dantegd) URL: #6354
PRs being backported: - [x] #6234 - [x] #6306 - [x] #6320 - [x] #6319 - [x] #6327 - [x] #6333 - [x] #6142 - [x] #6223 - [x] #6235 - [x] #6317 - [x] #6331 - [x] #6326 - [x] #6332 - [x] #6347 - [x] #6348 - [x] #6337 - [x] #6355 - [x] #6354 - [x] #6322 - [x] #6353 - [x] #6359 - [x] #6364 - [x] #6363 - [x] [FIL BATCH_TREE_REORG fix for SM90, 100 and 120](a3e419a) --------- Co-authored-by: William Hicks <[email protected]>
Support for zero code change acceleration. With the accelerator turned on, the scikit-learn example
./examples/miscellaneous/plot_kernel_ridge_regression.pyfails because it returns a cupy array fromThis fixes this.