Mahoutというか，その中にはそれっぽいライブラリが見つからなかったのだけど，科学技術計算の実験系で人工データを作りたいときに困ったので作ってみた．
import java.io.Serializable;
import java.util.Random;
import org.apache.mahout.math.DenseVector;
import org.apache.mahout.math.Vector;
public class SyntheticDataGenerator implements Serializable {
private static final long serialVersionUID = 1L;
private Random[] randoms;
private double[] means;
private double[] stdevs;
public SyntheticDataGenerator(long seed) {
means = new double[1];
means[0] = 0;
stdevs = new double[1];
stdevs[0] = 1;
init(seed, 1, means, stdevs);
}
public SyntheticDataGenerator(long seed, int cardinality, double[] means,
double[] stdevs) {
init(seed, cardinality, means, stdevs);
}
private void init(long seed, int cardinality, double[] means,
double[] stdevs) {
if (cardinality != means.length || cardinality != stdevs.length) {
throw new IllegalArgumentException("Invalid cardinality.");
}
randoms = new Random[cardinality];
for (int i = 0; i < cardinality; i++) {
randoms[i] = new Random(seed + i);
}
this.means = means;
this.stdevs = stdevs;
}
public double nextDouble() {
return nextDouble(0);
}
protected double nextDouble(int i) {
return randoms[i].nextGaussian() * stdevs[i] + means[i];
}
public double[] nextDoubles() {
double[] values = new double[randoms.length];
for (int i = 0; i < randoms.length; i++) {
values[i] = nextDouble(i);
}
return values;
}
public Vector nextVector() {
return new DenseVector(nextDoubles());
}
}
1次元の正規分布に基づく人工データを作りたいときには以下な感じ．
double[] means = new double[1]; double[] stdevs = new double[1]; means[0] = 10; // 平均 10 stdevs[0] = 5; // 標準偏差 5 SyntheticDataGenerator generator = new SyntheticDataGenerator(0, 1, means, stdevs);
あとは，generator.nextDouble() で値を取得していくと指定した分布の乱数が取得できる．そんで，多次元のデータが欲しい場合は，各次元ごとのmeansとstdevsを配列に格納して，generator.nextDoubles() としてやれば配列がとれるし，nextVector()でMahoutのVectorとして取得できる．