7_description.markdown

XGBoost

XgboostSample.java

import org.dmlc.xgboost4j.Booster;
import org.dmlc.xgboost4j.DMatrix;
import org.dmlc.xgboost4j.util.Trainer;
import org.dmlc.xgboost4j.util.XGBoostError;

import java.util.*;

public class XgboostSample {
    public static void main(String[] args) throws XGBoostError {
        // 学習データ
        DMatrix trainMat = new DMatrix("/tmp/iris.train.scale");
        // テストデータ
        DMatrix testMat = new DMatrix("/tmp/iris.test.scale");

        // パラメータの設定
        Map<String, Object> param = new HashMap<String, Object>() {
            {
                put("objective", "multi:softmax");
                put("num_class", 3);
            }
        };

        List<Map.Entry<String, DMatrix>> watchs =  new ArrayList<>();
        watchs.add(new AbstractMap.SimpleEntry<>("train", trainMat));
        watchs.add(new AbstractMap.SimpleEntry<>("test", testMat));

        // モデルの構築
        int round = 10;
        Booster booster = Trainer.train(param.entrySet(), trainMat, round, watchs, null, null);
        // 予測の実行
        float[][] predicts = booster.predict(testMat);

        System.out.println("predict length1: " + predicts.length);
        System.out.println("predict length2: " + predicts[0].length);
        System.out.println("error of predicts: " + eval(predicts, testMat));
    }

    private static float eval(float[][] predicts, DMatrix dmat) {
        float error = 0f;
        float[] labels;
        try {
            labels = dmat.getLabel();
        } catch (XGBoostError ex) {
            return -1f;
        }
        int nrow = predicts.length;
        for(int i=0; i<nrow; i++) {
            if(labels[i] != predicts[i][0]) {
                error++;
            }
        }
        System.out.println("error of count: " + error);
        return error/labels.length;
    }

}