@callstack/reassure-measure/lib/module/outlier-helpers.js

Performance measurement library for React and React Native

/* Adapted from https://github.com/sharkdp/hyperfine/blob/3b0918511aee4d6f8860bb663cb7a7af57bc3814/src/outlier_detection.rs */

import * as math from 'mathjs';

// Minimum modified Z-score for a datapoint to be an outlier. Here, 1.4826 is a factor that
// converts the MAD to an estimator for the standard deviation. The second factor is the number
// of standard deviations.
const OUTLIER_THRESHOLD = 1.4826 * 10;
export function findOutliers(items) {
  if (items.length <= 1) {
    return {
      results: items,
      outliers: []
    };
  }
  const durations = items.map(({
    duration
  }) => duration);

  // Compute the sample median and median absolute deviation (MAD)
  const median = math.median(durations);
  const mad = math.mad(durations);
  return items.reduce((acc, result) => {
    const modifiedZScore = (result.duration - median) / (mad > 0 ? mad : Number.EPSILON);

    // An outlier is a point that is larger than the modified Z-score threshold
    if (Math.abs(modifiedZScore) > OUTLIER_THRESHOLD) {
      acc.outliers.push(result);
    } else {
      acc.results.push(result);
    }
    return acc;
  }, {
    results: [],
    outliers: []
  });
}
//# sourceMappingURL=outlier-helpers.js.map