MNT replace Cython loss functions in SGD part 3 #28037
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mainly name changes: - loss(..) -> cy_loss(..) - dloss(..) -> cy_gradient(..)
|
I ran the asv benchmark for | Change | Before <main> | After <replace_sgd_with_common_loss_part_3> | Ratio | Benchmark (Parameter) |
|----------|----------------------------|----------------------------------------------------------|---------|---------------------------------------------------------------------------------|
| | 108M | 108M | 1 | linear_model.LogisticRegressionBenchmark.peakmem_fit('dense', 'saga', 1) |
| | 138M | 129M | 0.94 | linear_model.LogisticRegressionBenchmark.peakmem_fit('sparse', 'saga', 1) |
| | 113M | 115M | 1.02 | linear_model.LogisticRegressionBenchmark.peakmem_predict('dense', 'saga', 1) |
| | 115M | 116M | 1.01 | linear_model.LogisticRegressionBenchmark.peakmem_predict('sparse', 'saga', 1) |
| | 1.03±0s | 1.02±0.01s | 0.99 | linear_model.LogisticRegressionBenchmark.time_fit('dense', 'saga', 1) |
| | 984±1ms | 980±5ms | 1 | linear_model.LogisticRegressionBenchmark.time_fit('sparse', 'saga', 1) |
| + | 1.96±0.4ms | 2.16±0.3ms | 1.1 | linear_model.LogisticRegressionBenchmark.time_predict('dense', 'saga', 1) |
| | 2.41±0.07ms | 2.38±0.02ms | 0.99 | linear_model.LogisticRegressionBenchmark.time_predict('sparse', 'saga', 1) |
| | 0.7796506886650254 | 0.7796506886650254 | 1 | linear_model.LogisticRegressionBenchmark.track_test_score('dense', 'saga', 1) |
| | 0.5765140080078162 | 0.5765140080078162 | 1 | linear_model.LogisticRegressionBenchmark.track_test_score('sparse', 'saga', 1) |
| | 0.7992710685860992 | 0.7992710685860992 | 1 | linear_model.LogisticRegressionBenchmark.track_train_score('dense', 'saga', 1) |
| | 0.6908414295256007 | 0.6908414295256007 | 1 | linear_model.LogisticRegressionBenchmark.track_train_score('sparse', 'saga', 1) |I am not sure about this peculiar case where |
|
Here is another run which shows no change | Change | Before [main] | After [replace_sgd_with_common_loss_part_3] | Ratio | Benchmark (Parameter) |
|----------|----------------------------|----------------------------------------------------------|---------|---------------------------------------------------------------------------------|
| | 109M | 109M | 1 | linear_model.LogisticRegressionBenchmark.peakmem_fit('dense', 'saga', 1) |
| | 129M | 138M | 1.07 | linear_model.LogisticRegressionBenchmark.peakmem_fit('sparse', 'saga', 1) |
| | 113M | 113M | 1 | linear_model.LogisticRegressionBenchmark.peakmem_predict('dense', 'saga', 1) |
| | 116M | 114M | 0.99 | linear_model.LogisticRegressionBenchmark.peakmem_predict('sparse', 'saga', 1) |
| | 1.03±0.01s | 1.02±0.01s | 0.99 | linear_model.LogisticRegressionBenchmark.time_fit('dense', 'saga', 1) |
| | 984±2ms | 994±20ms | 1.01 | linear_model.LogisticRegressionBenchmark.time_fit('sparse', 'saga', 1) |
| | 2.08±0.3ms | 2.00±0.3ms | 0.96 | linear_model.LogisticRegressionBenchmark.time_predict('dense', 'saga', 1) |
| | 2.36±0.02ms | 2.38±0.05ms | 1.01 | linear_model.LogisticRegressionBenchmark.time_predict('sparse', 'saga', 1) |
| | 0.7796506886650254 | 0.7796506886650254 | 1 | linear_model.LogisticRegressionBenchmark.track_test_score('dense', 'saga', 1) |
| | 0.5765140080078162 | 0.5765140080078162 | 1 | linear_model.LogisticRegressionBenchmark.track_test_score('sparse', 'saga', 1) |
| | 0.7992710685860992 | 0.7992710685860992 | 1 | linear_model.LogisticRegressionBenchmark.track_train_score('dense', 'saga', 1) |
| | 0.6908414295256007 | 0.6908414295256007 | 1 | linear_model.LogisticRegressionBenchmark.track_train_score('sparse', 'saga', 1) |A further run shows that the performance has increased 😄 | Change | Before [156ef1b7] <main> | After [0d4fbe5c] <replace_sgd_with_common_loss_part_3> | Ratio | Benchmark (Parameter) |
|----------|----------------------------|----------------------------------------------------------|---------|---------------------------------------------------------------------------------|
| | 109M | 108M | 0.99 | linear_model.LogisticRegressionBenchmark.peakmem_fit('dense', 'saga', 1) |
| | 129M | 142M | 1.1 | linear_model.LogisticRegressionBenchmark.peakmem_fit('sparse', 'saga', 1) |
| | 113M | 116M | 1.03 | linear_model.LogisticRegressionBenchmark.peakmem_predict('dense', 'saga', 1) |
| | 115M | 115M | 1.01 | linear_model.LogisticRegressionBenchmark.peakmem_predict('sparse', 'saga', 1) |
| | 1.03±0s | 1.02±0.01s | 0.99 | linear_model.LogisticRegressionBenchmark.time_fit('dense', 'saga', 1) |
| | 986±10ms | 978±2ms | 0.99 | linear_model.LogisticRegressionBenchmark.time_fit('sparse', 'saga', 1) |
| - | 2.65±0.4ms | 2.25±0.3ms | 0.85 | linear_model.LogisticRegressionBenchmark.time_predict('dense', 'saga', 1) |
| | 2.34±0.01ms | 2.36±0.03ms | 1.01 | linear_model.LogisticRegressionBenchmark.time_predict('sparse', 'saga', 1) |
| | 0.7796506886650254 | 0.7796506886650254 | 1 | linear_model.LogisticRegressionBenchmark.track_test_score('dense', 'saga', 1) |
| | 0.5765140080078162 | 0.5765140080078162 | 1 | linear_model.LogisticRegressionBenchmark.track_test_score('sparse', 'saga', 1) |
| | 0.7992710685860992 | 0.7992710685860992 | 1 | linear_model.LogisticRegressionBenchmark.track_train_score('dense', 'saga', 1) |
| | 0.6908414295256007 | 0.6908414295256007 | 1 | linear_model.LogisticRegressionBenchmark.track_train_score('sparse', 'saga', 1) | |
OmarManzoor
approved these changes
Jul 24, 2024
jjerphan
approved these changes
Jul 24, 2024
Comment on lines
+94
to
+101
| cdef class CyHalfMultinomialLoss(): | ||
| cdef void cy_gradient( | ||
| self, | ||
| const floating_in y_true, | ||
| const floating_in[::1] raw_prediction, | ||
| const floating_in sample_weight, | ||
| floating_out[::1] gradient_out, | ||
| ) noexcept nogil |
There was a problem hiding this comment.
It is much nicer to be able to use fused types directly!
|
@jjerphan Once again thank you for the review and I leave the merging part to you. |
MarcBresson
pushed a commit
to MarcBresson/scikit-learn
that referenced
this pull request
Sep 2, 2024
Co-authored-by: Omar Salman <omar.salman@arbisoft.com> Co-authored-by: Omar Salman <omar.salman2007@gmail.com> Co-authored-by: Omar Salman <omar.salman@arbisoft>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Reference Issues/PRs
Follow-up of #28029 (which needs to be merged first), partially addresses #15123.
What does this implement/fix? Explain your changes.
This PR replaces the multinomial loss for SAGA.
Any other comments?
Only merge after release 1.5, i.e. this PR is to be released with v1.6.