001/*
002 * Copyright (C) 2012 The Guava Authors
003 *
004 * Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except
005 * in compliance with the License. You may obtain a copy of the License at
006 *
007 * http://www.apache.org/licenses/LICENSE-2.0
008 *
009 * Unless required by applicable law or agreed to in writing, software distributed under the License
010 * is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
011 * or implied. See the License for the specific language governing permissions and limitations under
012 * the License.
013 */
014
015package com.google.common.math;
016
017import static com.google.common.base.Preconditions.checkState;
018import static com.google.common.primitives.Doubles.isFinite;
019import static java.lang.Double.NaN;
020import static java.lang.Double.isNaN;
021
022import com.google.common.annotations.GwtIncompatible;
023import com.google.common.annotations.J2ktIncompatible;
024import com.google.common.primitives.Doubles;
025
026/**
027 * A mutable object which accumulates paired double values (e.g. points on a plane) and tracks some
028 * basic statistics over all the values added so far. This class is not thread safe.
029 *
030 * @author Pete Gillin
031 * @since 20.0
032 */
033@J2ktIncompatible
034@GwtIncompatible
035@ElementTypesAreNonnullByDefault
036public final class PairedStatsAccumulator {
037
038  // These fields must satisfy the requirements of PairedStats' constructor as well as those of the
039  // stat methods of this class.
040  private final StatsAccumulator xStats = new StatsAccumulator();
041  private final StatsAccumulator yStats = new StatsAccumulator();
042  private double sumOfProductsOfDeltas = 0.0;
043
044  /** Adds the given pair of values to the dataset. */
045  public void add(double x, double y) {
046    // We extend the recursive expression for the one-variable case at Art of Computer Programming
047    // vol. 2, Knuth, 4.2.2, (16) to the two-variable case. We have two value series x_i and y_i.
048    // We define the arithmetic means X_n = 1/n \sum_{i=1}^n x_i, and Y_n = 1/n \sum_{i=1}^n y_i.
049    // We also define the sum of the products of the differences from the means
050    //           C_n = \sum_{i=1}^n x_i y_i - n X_n Y_n
051    // for all n >= 1. Then for all n > 1:
052    //       C_{n-1} = \sum_{i=1}^{n-1} x_i y_i - (n-1) X_{n-1} Y_{n-1}
053    // C_n - C_{n-1} = x_n y_n - n X_n Y_n + (n-1) X_{n-1} Y_{n-1}
054    //               = x_n y_n - X_n [ y_n + (n-1) Y_{n-1} ] + [ n X_n - x_n ] Y_{n-1}
055    //               = x_n y_n - X_n y_n - x_n Y_{n-1} + X_n Y_{n-1}
056    //               = (x_n - X_n) (y_n - Y_{n-1})
057    xStats.add(x);
058    if (isFinite(x) && isFinite(y)) {
059      if (xStats.count() > 1) {
060        sumOfProductsOfDeltas += (x - xStats.mean()) * (y - yStats.mean());
061      }
062    } else {
063      sumOfProductsOfDeltas = NaN;
064    }
065    yStats.add(y);
066  }
067
068  /**
069   * Adds the given statistics to the dataset, as if the individual values used to compute the
070   * statistics had been added directly.
071   */
072  public void addAll(PairedStats values) {
073    if (values.count() == 0) {
074      return;
075    }
076
077    xStats.addAll(values.xStats());
078    if (yStats.count() == 0) {
079      sumOfProductsOfDeltas = values.sumOfProductsOfDeltas();
080    } else {
081      // This is a generalized version of the calculation in add(double, double) above. Note that
082      // non-finite inputs will have sumOfProductsOfDeltas = NaN, so non-finite values will result
083      // in NaN naturally.
084      sumOfProductsOfDeltas +=
085          values.sumOfProductsOfDeltas()
086              + (values.xStats().mean() - xStats.mean())
087                  * (values.yStats().mean() - yStats.mean())
088                  * values.count();
089    }
090    yStats.addAll(values.yStats());
091  }
092
093  /** Returns an immutable snapshot of the current statistics. */
094  public PairedStats snapshot() {
095    return new PairedStats(xStats.snapshot(), yStats.snapshot(), sumOfProductsOfDeltas);
096  }
097
098  /** Returns the number of pairs in the dataset. */
099  public long count() {
100    return xStats.count();
101  }
102
103  /** Returns an immutable snapshot of the statistics on the {@code x} values alone. */
104  public Stats xStats() {
105    return xStats.snapshot();
106  }
107
108  /** Returns an immutable snapshot of the statistics on the {@code y} values alone. */
109  public Stats yStats() {
110    return yStats.snapshot();
111  }
112
113  /**
114   * Returns the population covariance of the values. The count must be non-zero.
115   *
116   * <p>This is guaranteed to return zero if the dataset contains a single pair of finite values. It
117   * is not guaranteed to return zero when the dataset consists of the same pair of values multiple
118   * times, due to numerical errors.
119   *
120   * <h3>Non-finite values</h3>
121   *
122   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link
123   * Double#NEGATIVE_INFINITY}, or {@link Double#NaN}) then the result is {@link Double#NaN}.
124   *
125   * @throws IllegalStateException if the dataset is empty
126   */
127  public double populationCovariance() {
128    checkState(count() != 0);
129    return sumOfProductsOfDeltas / count();
130  }
131
132  /**
133   * Returns the sample covariance of the values. The count must be greater than one.
134   *
135   * <p>This is not guaranteed to return zero when the dataset consists of the same pair of values
136   * multiple times, due to numerical errors.
137   *
138   * <h3>Non-finite values</h3>
139   *
140   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link
141   * Double#NEGATIVE_INFINITY}, or {@link Double#NaN}) then the result is {@link Double#NaN}.
142   *
143   * @throws IllegalStateException if the dataset is empty or contains a single pair of values
144   */
145  public final double sampleCovariance() {
146    checkState(count() > 1);
147    return sumOfProductsOfDeltas / (count() - 1);
148  }
149
150  /**
151   * Returns the <a href="http://mathworld.wolfram.com/CorrelationCoefficient.html">Pearson's or
152   * product-moment correlation coefficient</a> of the values. The count must greater than one, and
153   * the {@code x} and {@code y} values must both have non-zero population variance (i.e. {@code
154   * xStats().populationVariance() > 0.0 && yStats().populationVariance() > 0.0}). The result is not
155   * guaranteed to be exactly +/-1 even when the data are perfectly (anti-)correlated, due to
156   * numerical errors. However, it is guaranteed to be in the inclusive range [-1, +1].
157   *
158   * <h3>Non-finite values</h3>
159   *
160   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link
161   * Double#NEGATIVE_INFINITY}, or {@link Double#NaN}) then the result is {@link Double#NaN}.
162   *
163   * @throws IllegalStateException if the dataset is empty or contains a single pair of values, or
164   *     either the {@code x} and {@code y} dataset has zero population variance
165   */
166  public final double pearsonsCorrelationCoefficient() {
167    checkState(count() > 1);
168    if (isNaN(sumOfProductsOfDeltas)) {
169      return NaN;
170    }
171    double xSumOfSquaresOfDeltas = xStats.sumOfSquaresOfDeltas();
172    double ySumOfSquaresOfDeltas = yStats.sumOfSquaresOfDeltas();
173    checkState(xSumOfSquaresOfDeltas > 0.0);
174    checkState(ySumOfSquaresOfDeltas > 0.0);
175    // The product of two positive numbers can be zero if the multiplication underflowed. We
176    // force a positive value by effectively rounding up to MIN_VALUE.
177    double productOfSumsOfSquaresOfDeltas =
178        ensurePositive(xSumOfSquaresOfDeltas * ySumOfSquaresOfDeltas);
179    return ensureInUnitRange(sumOfProductsOfDeltas / Math.sqrt(productOfSumsOfSquaresOfDeltas));
180  }
181
182  /**
183   * Returns a linear transformation giving the best fit to the data according to <a
184   * href="http://mathworld.wolfram.com/LeastSquaresFitting.html">Ordinary Least Squares linear
185   * regression</a> of {@code y} as a function of {@code x}. The count must be greater than one, and
186   * either the {@code x} or {@code y} data must have a non-zero population variance (i.e. {@code
187   * xStats().populationVariance() > 0.0 || yStats().populationVariance() > 0.0}). The result is
188   * guaranteed to be horizontal if there is variance in the {@code x} data but not the {@code y}
189   * data, and vertical if there is variance in the {@code y} data but not the {@code x} data.
190   *
191   * <p>This fit minimizes the root-mean-square error in {@code y} as a function of {@code x}. This
192   * error is defined as the square root of the mean of the squares of the differences between the
193   * actual {@code y} values of the data and the values predicted by the fit for the {@code x}
194   * values (i.e. it is the square root of the mean of the squares of the vertical distances between
195   * the data points and the best fit line). For this fit, this error is a fraction {@code sqrt(1 -
196   * R*R)} of the population standard deviation of {@code y}, where {@code R} is the Pearson's
197   * correlation coefficient (as given by {@link #pearsonsCorrelationCoefficient()}).
198   *
199   * <p>The corresponding root-mean-square error in {@code x} as a function of {@code y} is a
200   * fraction {@code sqrt(1/(R*R) - 1)} of the population standard deviation of {@code x}. This fit
201   * does not normally minimize that error: to do that, you should swap the roles of {@code x} and
202   * {@code y}.
203   *
204   * <h3>Non-finite values</h3>
205   *
206   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link
207   * Double#NEGATIVE_INFINITY}, or {@link Double#NaN}) then the result is {@link
208   * LinearTransformation#forNaN()}.
209   *
210   * @throws IllegalStateException if the dataset is empty or contains a single pair of values, or
211   *     both the {@code x} and {@code y} dataset have zero population variance
212   */
213  public final LinearTransformation leastSquaresFit() {
214    checkState(count() > 1);
215    if (isNaN(sumOfProductsOfDeltas)) {
216      return LinearTransformation.forNaN();
217    }
218    double xSumOfSquaresOfDeltas = xStats.sumOfSquaresOfDeltas();
219    if (xSumOfSquaresOfDeltas > 0.0) {
220      if (yStats.sumOfSquaresOfDeltas() > 0.0) {
221        return LinearTransformation.mapping(xStats.mean(), yStats.mean())
222            .withSlope(sumOfProductsOfDeltas / xSumOfSquaresOfDeltas);
223      } else {
224        return LinearTransformation.horizontal(yStats.mean());
225      }
226    } else {
227      checkState(yStats.sumOfSquaresOfDeltas() > 0.0);
228      return LinearTransformation.vertical(xStats.mean());
229    }
230  }
231
232  private double ensurePositive(double value) {
233    if (value > 0.0) {
234      return value;
235    } else {
236      return Double.MIN_VALUE;
237    }
238  }
239
240  private static double ensureInUnitRange(double value) {
241    return Doubles.constrainToRange(value, -1.0, 1.0);
242  }
243}