Model Api

fit!(model, X) ; fit!(model, X, y)

Trains model on the input data X and y (for supervised learning) or on just X (for unsupervised learning). The model object is always returned, allowing code like classifier = fit!(LogisticRegression(), X, y)

partial_fit!(model, X) ; partial_fit!(model, X, y)

Incrementally trains model on the new data X and y. For instance, this might perform a stochastic gradient descent step.

predict(model, X) returns the predicted class of each row in X (for classifiers) or the predicted value (for regressors).

predict_proba(model, X) returns an (N, C) matrix containing the probability that the n-th sample belongs to the c-th class. Call get_classes(model) to get the ordering of the classes.

predict_log_proba(model, X) is equivalent to log(predict_proba(model, X)) but can be more accurate (for small probabilities) and faster (avoiding the exponential).

For unsupervised learning models and for preprocessing, transform(model, X) applies the transformation from model to X, and returns a similar array (same number of rows, possibly different number of columns).

For unsupervised learning models, get_components(model) returns the matrix of the latent space, in (n*components, n*features) form. For matrix factorization methods, this corresponds to the principal components or latent vectors.

fit_transform!(model, X) is equivalent to transform(fit!(model, X), X) but can sometimes be more efficient.

fit_predict!(model, X) is equivalent to predict(fit!(model, X), X) but can sometimes be more efficient.

inverse_transform(model, X) applies the inverse of the model transformation.

For probabilistic models, score_samples(model, X) evaluates the density model on X.

score(model, X) and score(model, X, y) assign a score to how likely X or y|X is given the learned model parameters. The higher this score is, the better the model. This is used for cross-validation.

decision_function(model, X) returns the distance of the samples to the decision boundary.