// CHECK-DAG: %[[EPS:.+]] = mhlo.constant dense<1.000000e+00> : tensor<256xf32>
  // CHECK-DAG: %[[MEAN:.+]] = "tf.Mean"(%arg0, %[[CST_AXIS]]) <{keep_dims = false}> : (tensor<3x4x256x6xf32>, tensor<3xi32>) -> tensor<256xf32>
  // CHECK-DAG: %[[MEAN_BCAST:.+]] = "mhlo.dynamic_broadcast_in_dim"(%[[MEAN]], %[[X_SHAPE]]) <{broadcast_dimensions = dense<2> : tensor<1xi64>}> : (tensor<256xf32>, tensor<4xindex>) -> tensor<3x4x256x6xf32>

        mean += (value - mean) / (index + 1);
      } else {
        mean = calculateNewMeanNonFinite(mean, value);
      }
    }
    return mean;
  }

  /**
   * Returns the <a href="http://en.wikipedia.org/wiki/Arithmetic_mean">arithmetic mean</a> of the
   * values. The count must be non-zero.
   *
   * <p>The definition of the mean is the same as {@link Stats#mean}.
   *

      }
    } else {
      count++;
      if (isFinite(value) && isFinite(mean)) {
        // Art of Computer Programming vol. 2, Knuth, 4.2.2, (15) and (16)
        double delta = value - mean;
        mean += delta / count;
        sumOfSquaresOfDeltas += delta * (value - mean);
      } else {
        mean = calculateNewMeanNonFinite(mean, value);
        sumOfSquaresOfDeltas = NaN;
      }
      min = Math.min(min, value);

      } else if (values.hasAnyNegativeInfinity()) {
        assertWithMessage("mean of " + values).that(mean).isNegativeInfinity();
      } else {
        assertWithMessage("mean of " + values)
            .that(mean)
            .isWithin(ALLOWED_ERROR)
            .of(MANY_VALUES_MEAN);
      }
    }

        mean += (value - mean) / (index + 1);
      } else {
        mean = calculateNewMeanNonFinite(mean, value);
      }
    }
    return mean;
  }

  /**
   * Returns the <a href="http://en.wikipedia.org/wiki/Arithmetic_mean">arithmetic mean</a> of the
   * values. The count must be non-zero.
   *
   * <p>The definition of the mean is the same as {@link Stats#mean}.
   *

      }
    } else {
      count++;
      if (isFinite(value) && isFinite(mean)) {
        // Art of Computer Programming vol. 2, Knuth, 4.2.2, (15) and (16)
        double delta = value - mean;
        mean += delta / count;
        sumOfSquaresOfDeltas += delta * (value - mean);
      } else {
        mean = calculateNewMeanNonFinite(mean, value);
        sumOfSquaresOfDeltas = NaN;
      }
      min = Math.min(min, value);

   */
  @Deprecated
  public static double mean(long... values) {
    checkArgument(values.length > 0, "Cannot take mean of 0 values");
    long count = 1;
    double mean = values[0];
    for (int index = 1; index < values.length; ++index) {
      count++;
      // Art of Computer Programming vol. 2, Knuth, 4.2.2, (15)
      mean += (values[index] - mean) / count;
    }
    return mean;
  }

  /**

   */
  @Deprecated
  public static double mean(long... values) {
    checkArgument(values.length > 0, "Cannot take mean of 0 values");
    long count = 1;
    double mean = values[0];
    for (int index = 1; index < values.length; ++index) {
      count++;
      // Art of Computer Programming vol. 2, Knuth, 4.2.2, (15)
      mean += (values[index] - mean) / count;
    }
    return mean;
  }

  /**

        twoValuesAccumulator.xStats().mean(),
        twoValuesAccumulator.yStats().mean(),
        twoValuesAccumulator.xStats().populationVariance(),
        twoValuesAccumulator.populationCovariance());
    assertDiagonalLinearTransformation(
        twoValuesAccumulatorByAddAllPartitionedPairedStats.leastSquaresFit(),
        twoValuesAccumulatorByAddAllPartitionedPairedStats.xStats().mean(),

// checkSimilarDistribution returns success if the mean and stddev of the
// two statsResults are similar.
func (sr *statsResults) checkSimilarDistribution(expected *statsResults) error {
	if !nearEqual(sr.mean, expected.mean, expected.closeEnough, expected.maxError) {
		s := fmt.Sprintf("mean %v != %v (allowed error %v, %v)", sr.mean, expected.mean, expected.closeEnough, expected.maxError)
		fmt.Println(s)
		return errors.New(s)