GY-63_MS5611/libraries/Statistic

Statistic

Header-only statistic library for Arduino includes sum, average, variance and standard deviation.

The statistic::Statistic<T, C, bool _useStdDev> class template accepts 3 arguments:

• typename T: The floating point type used to represent the statistics.
• typename C: The unsigned integer type to store the number of values.
• typename _useStdDev: Compile-time flag for using variance and standard deviation.

To maintain backwards compatibility with API <= 0.4.4, the Statistic class implementation has been moved to the statistic namespace and a typedef statistic::Statistic<float, uint32_t, true> Statistic type definition has been created at global scope.

The useStdDev boolean was moved from a run-time to a compile-time option for two reasons. First, the compile-time option allows the optimizer to eliminate dead code (calculating standard deviation and variances) for a slightly smaller code size. Second, it was observed in uses of the library that the useStdDev boolean was set once in the class constructor and was never modified at run-time.

Description

The statistic library is made to get basic statistical information from a one dimensional set of data, e.g. a stream of values of a sensor.

The stability of the formulas is improved by the help of Gil Ross (Thanks!).

The template version (1.0.0) is created by Glen Cornell (Thanks!).

Related

• https://github.com/RobTillaart/Correlation
• https://github.com/RobTillaart/GST - Golden standard test metrics
• https://github.com/RobTillaart/RunningAngle
• https://github.com/RobTillaart/RunningAverage
• https://github.com/RobTillaart/RunningMedian
• https://github.com/RobTillaart/statHelpers - combinations & permutations
• https://github.com/RobTillaart/Statistic

Interface

#include "Statistic.h"

Constructor

• Statistic(void) Default constructor.
• statistic::Statistic<float, uint32_t, true> Constructor, with value type, count type, and standard deviation flag. The types mentioned are the defaults of the template. You can override e.g. statistic::Statistic<double, uint64_t, false> for many high precision values. (assumes double >> float).
• void clear() resets all internal variables and counters.

Core

• typename T add(const typename T value) returns value actually added to internal sum. If this differs from what should have been added, or even zero, the internal administration is running out of precision. If this happens after a lot of add() calls, it might become time to call clear(). Alternatively one need to define the statistic object with a more precise data type (typical double instead of float).
• typename C count() returns zero if count == zero (of course). Must be checked to interpret other values.
• typename T sum() returns zero if count == zero.
• typename T minimum() returns zero if count == zero.
• typename T maximum() returns zero if count == zero.
• typename T average() returns NAN if count == zero.

These three functions only work if useStdDev == true (in the template).

• typename T variance() returns NAN if count == zero.
• typename T pop_stdev() returns NAN if count == zero. pop_stdev = population standard deviation,
• typename T unbiased_stdev() returns NAN if count == zero.

Deprecated methods:

• Statistic(bool) Constructor previously used to enable/disable the standard deviation functions. This argument now has no effect. It is recommended to migrate your code to the default constructor (which now also implicitly calls clear()).
• void clear(bool) resets all variables. The boolean argument is ignored. It is recommended to migrate your code to clear() (with no arguments).

Operational

See examples.

Faq

See https://github.com/RobTillaart/Statistic/blob/master/FAQ.md

Future

Must

• update documentation
  • links that explain statistics in more depth

Should

• return values of sum(), minimum(), maximum() when count() == zero
  • should these be NaN, which is technically more correct?
  • does it exist for all value types? => No!
  • for now user responsibility to check count() first.
  • refactor _cnt to _count
• remove deprecated methods. (1.1.0)

Could

• add expected average EA compensation trick
  • every add will subtract EA before added to sum,
  • this will keep the _sum to around zero.
  • this will move average() to around zero.
  • do not forget to add EA to average.
  • do not forget to add EA times count for sum.
  • does not affect the std_dev()
  • all functions will become slightly slower.

Wont