public double statistic()

in commons-statistics-inference/src/main/java/org/apache/commons/statistics/inference/GTest.java [194:249]

• 36 lines of code
• 7 McCabe index (conditional complexity)

    public double statistic(long[][] counts) {
        Arguments.checkCategoriesRequiredSize(counts.length, 2);
        Arguments.checkValuesRequiredSize(counts[0].length, 2);
        Arguments.checkRectangular(counts);
        Arguments.checkNonNegative(counts);

        final int ni = counts.length;
        final int nj = counts[0].length;

        // Compute row, column and total sums
        final double[] sumi = new double[ni];
        final double[] sumj = new double[nj];
        double n = 0;
        // We can sum data on the first pass. See below for computation details.
        final Sum sum = Sum.create();
        for (int i = 0; i < ni; i++) {
            for (int j = 0; j < nj; j++) {
                final long c = counts[i][j];
                sumi[i] += c;
                sumj[j] += c;
                if (c > 1) {
                    sum.add(c * Math.log(c));
                }
            }
            checkNonZero(sumi[i], "Row", i);
            n += sumi[i];
        }

        for (int j = 0; j < nj; j++) {
            checkNonZero(sumj[j], "Column", j);
        }

        // This computes a modified form of the Shannon entropy H without requiring
        // normalisation of observations to probabilities and without negation,
        // i.e. we compute n * [ H(r) + H(c) - H(r,c) ] as [ H'(r,c) - H'(r) - H'(c) ].

        // H  = -sum (p * log(p))
        // H' = n * sum (p * log(p))
        //    = n * sum (o/n * log(o/n))
        //    = n * [ sum(o/n * log(o)) - sum(o/n * log(n)) ]
        //    = sum(o * log(o)) - n log(n)

        // After 3 modified entropy sums H'(r,c) - H'(r) - H'(c) compensation is (-1 + 2) * n log(n)
        sum.addProduct(n, Math.log(n));
        // Negative terms
        final Sum sum2 = Sum.create();
        // All these counts are above zero so no check for zeros
        for (final double c : sumi) {
            sum2.add(c * -Math.log(c));
        }
        for (final double c : sumj) {
            sum2.add(c * -Math.log(c));
        }

        return sum.add(sum2).getAsDouble() * 2;
    }