Train a supervised match combiner — np

Fits a classifier that maps the per-field similarity vector to a match probability — the model consumed by np_score() method = "model". The built-in engine is logistic regression; any engine returning an object with a predict(obj, newdata, type = "response") method can be supplied.

Usage

np_train(pairs, engine = "logit", features = attr(pairs, "cmp_cols"))

Arguments

pairs: A labelled np_pairs frame (see np_label()); rows with NA label are ignored.
engine: "logit" (default) or a function (x, y) returning a fitted model, where x is the similarity data frame and y is 0/1.
features: Similarity columns to use. Defaults to the compared fields.

Value

An np_model object usable as np_score(..., model = ).