Interfacing with Sklearn supervised models

[tlienart]

This issue is more to keep track of things than to ask for help

https://scikit-learn.org/stable/modules/classes.html

Supervised models

Dummy models

?	Name	Has predict_proba?	comment
✅	DummyClassifier	yes
✅	DummyRegressor	no

Generalized Linear Models

https://scikit-learn.org/stable/modules/classes.html#module-sklearn.linear_model

?	Name	Has predict_proba?	comment
✅	ARDRegression	no
✅	BayesianRidge	no
✅	ElasticNet	no	1
✅	ElasticNetCV	no	1
✅	HuberRegressor	no
✅	Lars	no	1
✅	LarsCV	no	1
✅	Lasso	no	1
✅	LassoCV	no	1
✅	LassoLars	no	1
✅	LassoLarsCV	no	1
✅	LassoLarsIC	no	1
✅	LinearRegression	no
✅	LogisticRegression	yes	2
✅	LogisticRegressionCV	yes	2
✅	MultiTaskLasso	no	1
✅	MultiTaskElasticNet	no	1
✅	MultiTaskLassoCV	no	1
✅	MultiTaskElasticNetCV	no	1
✅	OrthogonalMatchingPursuit	no
✅	OrthogonalMatchingPursuitCV	no
✅	PassiveAggressiveClassifier	no
✅	PassiveAggressiveRegressor	no
✅	Perceptron	no
✅	RANSACRegressor	no
✅	Ridge	no
✅	RidgeClassifier	no
✅	RidgeClassifierCV	no
✅	RidgeCV	no
✅	SGDClassifier	yes
✅	SGDRegressor	no
✅	TheilSenRegressor	no

1. could extract path
2. there's also predict_log_proba which we may want to interface with (?)

Gaussian Processes

?	Name	Has predict_proba?	comment
✅	GPClassifier	yes
✅	GPRegressor	no

Ensemble models

?	Name	Has predict_proba?	comment
✅	AdaboostClassif	yes
✅	AdaboostReg	no
✅	BaggingClassif	yes
✅	BaggingReg	no
✅	ExtraTreesClassif	yes
✅	ExtraTreesReg	no
✅	GDBClassif	yes
✅	GDBReg	no
✅	RFClassif	yes
✅	RFReg	no
❌	VotingClassif	no
❌	VotingReg	no
❌	HGBClassif	yes	experimental
❌	HGBReg	no	experimental

Simple models

?	Name	Has predict_proba?	comment
✅	BernoulliNB	yes
✅	GaussianNB	yes
✅	MultinomialNB	yes
✅	ComplementNB	yes
✅	KNeighborsClassifier	yes
✅	KNeighborsRegressor	no
❌	RadiusNeighborsClassifier	no
❌	RadiusNeighborsRegressor	no
❌	NearestCentroid (classif)	no

NNs

?	Name	Has predict_proba?	comment
❌	MLPClassifier	yes
❌	MLPRegressor	no

SVMs

?	Name	Has predict_proba?	comment
✅	LinearSVC	no
✅	LinearSVR	no
✅	NuSVC	no
✅	NuSVR	no
✅	SVC	no
✅	SVR	no

Exotic models

?	Name	Has predict_proba?	comment
❌	IsotonicRegression	no
❌	KernelRidgeRegression	no
❌	BayesianGaussianMixture	yes
❌	GaussianMixture	yes
❌	CalibratedClassifierCV	yes

[tlienart]

Hmm there's also LDA and QDA I should add

• QDA
• LDA

[tlienart]

Probably also those: https://scikit-learn.org/stable/modules/classes.html#module-sklearn.cross_decomposition

PLSR is used in ISL so could be good to complete that

[ablaom]

Not familiar with cross_decomposition. Can I think of this as a transformer, like we do PCA?