001/*
002 * Licensed to the Apache Software Foundation (ASF) under one or more
003 * contributor license agreements.  See the NOTICE file distributed with
004 * this work for additional information regarding copyright ownership.
005 * The ASF licenses this file to You under the Apache License, Version 2.0
006 * (the "License"); you may not use this file except in compliance with
007 * the License.  You may obtain a copy of the License at
008 *
009 *      http://www.apache.org/licenses/LICENSE-2.0
010 *
011 * Unless required by applicable law or agreed to in writing, software
012 * distributed under the License is distributed on an "AS IS" BASIS,
013 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
014 * See the License for the specific language governing permissions and
015 * limitations under the License.
016 */
017package org.apache.commons.statistics.descriptive;
018
019import java.math.BigInteger;
020
021/**
022 * Computes the variance of the available values. The default implementation uses the
023 * following definition of the <em>sample variance</em>:
024 *
025 * <p>\[ \tfrac{1}{n-1} \sum_{i=1}^n (x_i-\overline{x})^2 \]
026 *
027 * <p>where \( \overline{x} \) is the sample mean, and \( n \) is the number of samples.
028 *
029 * <ul>
030 *   <li>The result is {@code NaN} if no values are added.
031 *   <li>The result is zero if there is one value in the data set.
032 * </ul>
033 *
034 * <p>The use of the term \( n − 1 \) is called Bessel's correction. This is an unbiased
035 * estimator of the variance of a hypothetical infinite population. If the
036 * {@link #setBiased(boolean) biased} option is enabled the normalisation factor is
037 * changed to \( \frac{1}{n} \) for a biased estimator of the <em>sample variance</em>.
038 *
039 * <p>The implementation uses an exact integer sum to compute the scaled (by \( n \))
040 * sum of squared deviations from the mean; this is normalised by the scaled correction factor.
041 *
042 * <p>\[ \frac {n \times \sum_{i=1}^n x_i^2 - (\sum_{i=1}^n x_i)^2}{n \times (n - 1)} \]
043 *
044 * <p>Supports up to 2<sup>63</sup> (exclusive) observations.
045 * This implementation does not check for overflow of the count.
046 *
047 * <p>This class is designed to work with (though does not require)
048 * {@linkplain java.util.stream streams}.
049 *
050 * <p><strong>This implementation is not thread safe.</strong>
051 * If multiple threads access an instance of this class concurrently,
052 * and at least one of the threads invokes the {@link java.util.function.IntConsumer#accept(int) accept} or
053 * {@link StatisticAccumulator#combine(StatisticResult) combine} method, it must be synchronized externally.
054 *
055 * <p>However, it is safe to use {@link java.util.function.IntConsumer#accept(int) accept}
056 * and {@link StatisticAccumulator#combine(StatisticResult) combine}
057 * as {@code accumulator} and {@code combiner} functions of
058 * {@link java.util.stream.Collector Collector} on a parallel stream,
059 * because the parallel implementation of {@link java.util.stream.Stream#collect Stream.collect()}
060 * provides the necessary partitioning, isolation, and merging of results for
061 * safe and efficient parallel execution.
062 *
063 * @see <a href="https://en.wikipedia.org/wiki/variance">variance (Wikipedia)</a>
064 * @see <a href="https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance">
065 *   Algorithms for computing the variance (Wikipedia)</a>
066 * @see <a href="https://en.wikipedia.org/wiki/Bessel%27s_correction">Bessel&#39;s correction</a>
067 * @since 1.1
068 */
069public final class IntVariance implements IntStatistic, StatisticAccumulator<IntVariance> {
070    /** Small array sample size.
071     * Used to avoid computing with UInt96 then converting to UInt128. */
072    static final int SMALL_SAMPLE = 10;
073
074    /** Sum of the squared values. */
075    private final UInt128 sumSq;
076    /** Sum of the values. */
077    private final Int128 sum;
078    /** Count of values that have been added. */
079    private long n;
080
081    /** Flag to control if the statistic is biased, or should use a bias correction. */
082    private boolean biased;
083
084    /**
085     * Create an instance.
086     */
087    private IntVariance() {
088        this(UInt128.create(), Int128.create(), 0);
089    }
090
091    /**
092     * Create an instance.
093     *
094     * @param sumSq Sum of the squared values.
095     * @param sum Sum of the values.
096     * @param n Count of values that have been added.
097     */
098    private IntVariance(UInt128 sumSq, Int128 sum, int n) {
099        this.sumSq = sumSq;
100        this.sum = sum;
101        this.n = n;
102    }
103
104    /**
105     * Creates an instance.
106     *
107     * <p>The initial result is {@code NaN}.
108     *
109     * @return {@code IntVariance} instance.
110     */
111    public static IntVariance create() {
112        return new IntVariance();
113    }
114
115    /**
116     * Returns an instance populated using the input {@code values}.
117     *
118     * @param values Values.
119     * @return {@code IntVariance} instance.
120     */
121    public static IntVariance of(int... values) {
122        // Small arrays can be processed using the object
123        if (values.length < SMALL_SAMPLE) {
124            final IntVariance stat = new IntVariance();
125            for (final int x : values) {
126                stat.accept(x);
127            }
128            return stat;
129        }
130
131        // Arrays can be processed using specialised counts knowing the maximum limit
132        // for an array is 2^31 values.
133        long s = 0;
134        final UInt96 ss = UInt96.create();
135        // Process pairs as we know two maximum value int^2 will not overflow
136        // an unsigned long.
137        final int end = values.length & ~0x1;
138        for (int i = 0; i < end; i += 2) {
139            final long x = values[i];
140            final long y = values[i + 1];
141            s += x + y;
142            ss.addPositive(x * x + y * y);
143        }
144        if (end < values.length) {
145            final long x = values[end];
146            s += x;
147            ss.addPositive(x * x);
148        }
149
150        // Convert
151        return new IntVariance(UInt128.of(ss), Int128.of(s), values.length);
152    }
153
154    /**
155     * Updates the state of the statistic to reflect the addition of {@code value}.
156     *
157     * @param value Value.
158     */
159    @Override
160    public void accept(int value) {
161        sumSq.addPositive((long) value * value);
162        sum.add(value);
163        n++;
164    }
165
166    /**
167     * Gets the variance of all input values.
168     *
169     * <p>When no values have been added, the result is {@code NaN}.
170     *
171     * @return variance of all values.
172     */
173    @Override
174    public double getAsDouble() {
175        return computeVarianceOrStd(sumSq, sum, n, biased, false);
176    }
177
178    /**
179     * Compute the variance (or standard deviation).
180     *
181     * <p>The {@code std} flag controls if the result is returned as the standard deviation
182     * using the {@link Math#sqrt(double) square root} function.
183     *
184     * @param sumSq Sum of the squared values.
185     * @param sum Sum of the values.
186     * @param n Count of values that have been added.
187     * @param biased Flag to control if the statistic is biased, or should use a bias correction.
188     * @param std Flag to control if the statistic is the standard deviation.
189     * @return the variance (or standard deviation)
190     */
191    static double computeVarianceOrStd(UInt128 sumSq, Int128 sum, long n, boolean biased, boolean std) {
192        if (n == 0) {
193            return Double.NaN;
194        }
195        // Avoid a divide by zero
196        if (n == 1) {
197            return 0;
198        }
199        // Sum-of-squared deviations: sum(x^2) - sum(x)^2 / n
200        // Sum-of-squared deviations precursor: n * sum(x^2) - sum(x)^2
201        // The precursor is computed in integer precision.
202        // The divide uses double precision.
203        // This ensures we avoid cancellation in the difference and use a fast divide.
204        // The result is limited to by the rounding in the double computation.
205        final double diff = computeSSDevN(sumSq, sum, n);
206        final long n0 = biased ? n : n - 1;
207        final double v = diff / IntMath.unsignedMultiplyToDouble(n, n0);
208        if (std) {
209            return Math.sqrt(v);
210        }
211        return v;
212    }
213
214    /**
215     * Compute the sum-of-squared deviations multiplied by the count of values:
216     * {@code n * sum(x^2) - sum(x)^2}.
217     *
218     * @param sumSq Sum of the squared values.
219     * @param sum Sum of the values.
220     * @param n Count of values that have been added.
221     * @return the sum-of-squared deviations precursor
222     */
223    private static double computeSSDevN(UInt128 sumSq, Int128 sum, long n) {
224        // Compute the term if possible using fast integer arithmetic.
225        // 128-bit sum(x^2) * n will be OK when the upper 32-bits are zero.
226        // 128-bit sum(x)^2 will be OK when the upper 64-bits are zero.
227        // Both are safe when n < 2^32.
228        if ((n >>> Integer.SIZE) == 0) {
229            return sumSq.unsignedMultiply((int) n).subtract(sum.squareLow()).toDouble();
230        } else {
231            return sumSq.toBigInteger().multiply(BigInteger.valueOf(n))
232                .subtract(square(sum.toBigInteger())).doubleValue();
233        }
234    }
235
236    /**
237     * Compute the sum of the squared deviations from the mean.
238     *
239     * <p>This is a helper method used in higher order moments.
240     *
241     * @return the sum of the squared deviations
242     */
243    double computeSumOfSquaredDeviations() {
244        return computeSSDevN(sumSq, sum, n) / n;
245    }
246
247    /**
248     * Compute the mean.
249     *
250     * <p>This is a helper method used in higher order moments.
251     *
252     * @return the mean
253     */
254    double computeMean() {
255        return IntMean.computeMean(sum, n);
256    }
257
258    /**
259     * Convenience method to square a BigInteger.
260     *
261     * @param x Value
262     * @return x^2
263     */
264    private static BigInteger square(BigInteger x) {
265        return x.multiply(x);
266    }
267
268    @Override
269    public IntVariance combine(IntVariance other) {
270        sumSq.add(other.sumSq);
271        sum.add(other.sum);
272        n += other.n;
273        return this;
274    }
275
276    /**
277     * Sets the value of the biased flag. The default value is {@code false}.
278     *
279     * <p>If {@code false} the sum of squared deviations from the sample mean is normalised by
280     * {@code n - 1} where {@code n} is the number of samples. This is Bessel's correction
281     * for an unbiased estimator of the variance of a hypothetical infinite population.
282     *
283     * <p>If {@code true} the sum of squared deviations is normalised by the number of samples
284     * {@code n}.
285     *
286     * <p>Note: This option only applies when {@code n > 1}. The variance of {@code n = 1} is
287     * always 0.
288     *
289     * <p>This flag only controls the final computation of the statistic. The value of this flag
290     * will not affect compatibility between instances during a {@link #combine(IntVariance) combine}
291     * operation.
292     *
293     * @param v Value.
294     * @return {@code this} instance
295     */
296    public IntVariance setBiased(boolean v) {
297        biased = v;
298        return this;
299    }
300}