scikit-learn 1.0 is finally out! πππ
Here are my top 5 features from this release:
Thread π
Here are my top 5 features from this release:
Thread π
1. SplineTransformer:
Allows you to add nonlinear features without using pure polynomials. Splines are "piecewise polynomials" and have significant advantages over pure polynomials.
Notice how the spline (orange) can match the ground truth (black) better than the polynomials.
Allows you to add nonlinear features without using pure polynomials. Splines are "piecewise polynomials" and have significant advantages over pure polynomials.
Notice how the spline (orange) can match the ground truth (black) better than the polynomials.
2. SGDOneClassSVM:
Allows for unsupervised outlier detection using stochastic gradient descent.
Unlike OneClassSVM, it's suitable for large datasets.
More details:
scikit-learn.org/stable/modulesβ¦
Allows for unsupervised outlier detection using stochastic gradient descent.
Unlike OneClassSVM, it's suitable for large datasets.
More details:
scikit-learn.org/stable/modulesβ¦
3. HistGradientBoostingClassifier and Regressor:
These aren't new, but they're no longer experimental!
They have native support for missing values (no need to impute) and native support for categorical features (no need to one-hot encode).
More details:
These aren't new, but they're no longer experimental!
They have native support for missing values (no need to impute) and native support for categorical features (no need to one-hot encode).
More details:
https://twitter.com/justmarkham/status/1408013015530840064
4. More flexible API for plotting a confusion matrix, ROC curve, precision-recall curve, etc:
You can now generate these plots using predictions (my preferred option) OR an estimator (which used to be the only option).
More details:
You can now generate these plots using predictions (my preferred option) OR an estimator (which used to be the only option).
More details:
https://twitter.com/justmarkham/status/1415679379795963914
5. Most parameters are now required to be passed as keyword arguments (instead of positional arguments):
Requires you to write SVC(C=0.1, kernel='linear') instead of SVC(0.1, 'linear').
Makes code far more readable!
More details:
Requires you to write SVC(C=0.1, kernel='linear') instead of SVC(0.1, 'linear').
Makes code far more readable!
More details:
https://twitter.com/justmarkham/status/1436014153358053387
To be clear, scikit-learn has been stable for years, and releasing version 1.0 is just a way to indicate that to users.
Regardless, 1.0 is a HUGE milestone, and we should THANK the thousands of contributors who have donated countless hours to this project in the past decade! π
Regardless, 1.0 is a HUGE milestone, and we should THANK the thousands of contributors who have donated countless hours to this project in the past decade! π
Are you new to Machine Learning & scikit-learn? Get started with my FREE 4-hour course:
π courses.dataschool.io/introduction-tβ¦ π
Or do you want to learn ML best practices & improve your scikit-learn skills? Take my FREE 3-hour course:
π courses.dataschool.io/scikit-learn-tβ¦ π
π courses.dataschool.io/introduction-tβ¦ π
Or do you want to learn ML best practices & improve your scikit-learn skills? Take my FREE 3-hour course:
π courses.dataschool.io/scikit-learn-tβ¦ π
β’ β’ β’
Missing some Tweet in this thread? You can try to
force a refresh
γ
