[MRG] accelerate plot_gradient_boosting_regularization.py example #21598 #21611
Conversation
|
If the final output hasn't changed, we may be able to push further and speed up the example even more. Thanks for the work @sply88 |
|
Original figure in example looks like this: Could speed it up a bit more to around 9s by only using 400 boosting iterations. So the x-Axis of the figure in my original PR comment would end at 400 and the yellow and blue lines would not cross anymore. I don't think this would be a big issue because it would still be obvious that shrinkage is good and no-shrinkage (e.g. blue and yellow lines) is bad. |
| X_train, X_test = X[:1500], X[1500:] | ||
| y_train, y_test = y[:1500], y[1500:] |
There was a problem hiding this comment.
you could also try reducing the number of samples in the make_hastie_10_2 line above
There was a problem hiding this comment.
Have reduced both number of samples and number of estimators to get down to 5s.
Output below. Main message is still obvious I think.
| @@ -32,17 +32,17 @@ | |||
| from sklearn import datasets | |||
|
|
|||
|
|
|||
| X, y = datasets.make_hastie_10_2(n_samples=12000, random_state=1) | |||
| X, y = datasets.make_hastie_10_2(n_samples=3000, random_state=1) | |||
There was a problem hiding this comment.
What do you think of setting the following settings?
from sklearn.model_selection import train_test_split
X, y = datasets.make_hastie_10_2(n_samples=4000, random_state=1)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.8, random_state=0)
original_params = {
"n_estimators": 400,
...
}It looks like it keeps a very similar message as the original:
…t-learn#21598 (scikit-learn#21611) * accelerate plot_gradient_boosting_regularization.py example scikit-learn#21598 * speed up by less samples and less trees * use train_test_split instead of slicing
…t-learn#21598 (scikit-learn#21611) * accelerate plot_gradient_boosting_regularization.py example scikit-learn#21598 * speed up by less samples and less trees * use train_test_split instead of slicing
Speeds up ../examples/ensemble/plot_gradient_boosting_regularization.py (Issue #21598) by
n_estimatorsfrom 1000 to 600Reduction of
n_estimatorsis compensated by increasing the learning rate from 0.1 to 0.2 (for models with shrinkage).For me example runs in 13 sec now (previously plus 30).
Main message of final figure does not change:
