(ns lambda-ml.random-forest

"Random forest classification and regression learning.

Example usage:

```

(def data [[0 0 0] [0 1 1] [1 0 1] [1 1 0]])

(def fit

(let [n 1001

min-split 2

min-leaf 1

max-features 2]

(-> (make-random-forest-classifier n min-split min-leaf max-features)

(random-forest-fit data))))

(random-forest-predict fit (map butlast data))

;;=> (0 1 1 0)

```"

(:require [lambda-ml.decision-tree :as dt]

[lambda-ml.ensemble :as e]))

(def random-forest-fit e/bagging-ensemble-fit)

(def random-forest-predict e/bagging-ensemble-predict)

(defn make-random-forest-classifier

[n min-split min-leaf max-features]

(let [rate 1.0

estimator (dt/make-classification-tree dt/gini-impurity min-split min-leaf max-features)]

(-> #(e/add-bagging-estimator % estimator dt/decision-tree-fit dt/decision-tree-predict)

(iterate (e/make-bagging-classifier rate))

(nth n))))

(defn make-random-forest-regressor

[n min-split min-leaf max-features]

(let [rate 1.0

estimator (dt/make-regression-tree dt/mean-squared-error min-split min-leaf max-features)]

(-> #(e/add-bagging-estimator % estimator dt/decision-tree-fit dt/decision-tree-predict)

(iterate (e/make-bagging-regressor rate))

(nth n))))