tech.xgboost

XGBoost bindings for clojure.

• Datasets are sequences of maps
• Uses tech.datatype for efficient conversion of data into xgboost.

Example

  (:require [tech.xgboost :as xgboost]
            [tech.xgboost.dataset :as dataset]
            [tech.xgboost.util :as util])

  (let [f (partial * 2)
        observe (fn []
                  (let [x (- (* 20 (rand)) 10)
                        y (f x)]
                    {:x x :y y}))
        train-dataset (->> (repeatedly observe)
                           (take 1000))
        test-dataset (for [x (range -9.9 10 0.1)] {:x x})
        test-labels (map (comp f :x) test-dataset)
        model (xgboost/train train-dataset [:x] :y :regression {})
        test-output (xgboost/predict model test-dataset [:x])
        mse (util/MSE test-output test-labels)]
    (is (< mse 0.01)))


  (let [f (partial * 2)
        observe (fn []
                  (let [x (- (* 20 (rand)) 10)
                        y (f x)]
                    {:x x :y y}))
        dataset (->> (repeatedly observe)
                     (take 1000))
        feature-keys [:x]
        label :y
        train-fn #(xgboost/train % feature-keys label :regression {})
        predict-fn #(xgboost/predict %1 %2 feature-keys)
        mse (->> (dataset/dataset->k-fold-datasets dataset 10)
                 (dataset/average-prediction-error train-fn predict-fn
                                                   label util/MSE))]
    (is (< mse 0.01)))

License

Copyright © 2018 Tech Ascent, LLC

Distributed under the Eclipse Public License either version 1.0 or (at your option) any later version.