# criterion performance measurements

## overview

want to understand this report?

{{#report}}## {{name}}

lower bound | estimate | upper bound | |
---|---|---|---|

OLS regression | xxx | xxx | xxx |

R² goodness-of-fit | xxx | xxx | xxx |

Mean execution time | {{anMeanLowerBound}} | {{anMean.estPoint}} | {{anMeanUpperBound}} |

Standard deviation | {{anStdDevLowerBound}} | {{anStdDev.estPoint}} | {{anStdDevUpperBound}} |

Outlying measurements have {{anOutlierVar.ovDesc}} ({{anOutlierVar.ovFraction}}%) effect on estimated standard deviation.

{{/report}}## understanding this report

In this report, each function benchmarked by criterion is assigned a section of its own. The charts in each section are active; if you hover your mouse over data points and annotations, you will see more details.

- The chart on the left is a kernel density estimate (also known as a KDE) of time measurements. This graphs the probability of any given time measurement occurring. A spike indicates that a measurement of a particular time occurred; its height indicates how often that measurement was repeated.
- The chart on the right is the raw data from which the kernel
density estimate is built. The
*x*axis indicates the number of loop iterations, while the*y*axis shows measured execution time for the given number of loop iterations. The line behind the values is the linear regression prediction of execution time for a given number of iterations. Ideally, all measurements will be on (or very near) this line.

Under the charts is a small table. The first two rows are the results of a linear regression run on the measurements displayed in the right-hand chart.

*OLS regression*indicates the time estimated for a single loop iteration using an ordinary least-squares regression model. This number is more accurate than the*mean*estimate below it, as it more effectively eliminates measurement overhead and other constant factors.*R² goodness-of-fit*is a measure of how accurately the linear regression model fits the observed measurements. If the measurements are not too noisy, R² should lie between 0.99 and 1, indicating an excellent fit. If the number is below 0.99, something is confounding the accuracy of the linear model.*Mean execution time*and*standard deviation*are statistics calculated from execution time divided by number of iterations.

We use a statistical technique called the bootstrap to provide confidence intervals on our estimates. The bootstrap-derived upper and lower bounds on estimates let you see how accurate we believe those estimates to be. (Hover the mouse over the table headers to see the confidence levels.)

A noisy benchmarking environment can cause some or many measurements to fall far from the mean. These outlying measurements can have a significant inflationary effect on the estimate of the standard deviation. We calculate and display an estimate of the extent to which the standard deviation has been inflated by outliers.