You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

180 lines
4.9 KiB

#ifndef __CORESTATS_H__
#define __CORESTATS_H__
class Stats {
public:
virtual ~Stats() {}
/// Resets the statistics.
virtual void reset(float startMean=0, float startVar=0) = 0;
/// Adds a value to the statistics (returns the mean).
virtual float update(float value) = 0;
/// The statistics.
virtual float mean() const = 0;
virtual float var() const = 0;
virtual float stddev() const;
/// Returns the normalized value according to the computed statistics (mean and variance).
float normalize(float value) const;
};
class SimpleStats : public Stats {
public:
float _mean;
float _mean2; // mean of squared values
unsigned long _nSamples;
/// ctor
SimpleStats(float startMean=0, float startVar=0);
virtual ~SimpleStats() {}
/// Resets the statistics.
virtual void reset(float startMean=0, float startVar=0);
/// Adds a value to the statistics (returns the mean).
virtual float update(float value);
/// The statistics.
virtual float mean() const { return _mean; }
// The var() and stddev() are the population (ie. not the sample) variance and standard dev, so technically
// they should be readjusted by multiplying it by _nSamples / (_nSamples-1). But with a lot of samples the
// difference vanishes and we priviledged less floating points computations over precision.
virtual float var() const;
};
/// An exponential moving average class.
class MovingAverage {
public:
// The alpha (mixing) variable (in [0,1]).
float _alpha;
// The current value of the exponential moving average.
float _value;
/**
* Constructs the moving average, starting with #startValue# as its value. The #alphaOrN# argument
* has two options:
* - if <= 1 then it's used directly as the alpha value
* - if > 1 then it's used as the "number of items that are considered from the past" (*)
* (*) Of course this is an approximation. It actually sets the alpha value to 2 / (n - 1)
*/
MovingAverage(float alphaOrN=1);
MovingAverage(float alphaOrN, float startValue);
virtual ~MovingAverage() {}
/// Change the smoothing factor to #alphaOrN#.
void setAlphaOrN(float alphaOrN);
/// Resets the moving average.
void reset();
/// Resets the moving average to #startValue#.
void reset(float startValue);
/// Updates the moving average with new value #v# (also returns the current value).
float update(float v);
/// Returns the value of the moving average. This is undefined if isValid() == false.
float get() const { return _value; }
/// Returns true iff the moving average has already been started.
bool isStarted() const;
/// Returns the alpha value.
float alpha() const { return _alpha; }
protected:
void _setStarted(bool start);
};
class MovingStats : public Stats {
public:
MovingAverage avg;
float _var;
/**
* Constructs the moving statistics, starting with #startMean# and #startVar# as initial mean and
* variance. The #alphaOrN# argument has two options:
* - if <= 1 then it's used directly as the alpha value
* - if > 1 then it's used as the "number of items that are considered from the past" (*)
* (*) Of course this is an approximation. It actually sets the alpha value to 2 / (n - 1)
*/
MovingStats(float alphaOrN=1);
MovingStats(float alphaOrN, float startMean, float startVar);
virtual ~MovingStats() {}
/// Resets the statistics.
virtual void reset();
/// Resets the statistics.
virtual void reset(float startMean, float startVar);
/// Adds a value to the statistics (returns the mean).
virtual float update(float value);
/// The statistics.
virtual float mean() const { return avg.get(); }
virtual float var() const { return _var; }
virtual bool isStarted() const;
};
/// Adaptive normalizer: normalizes values on-the-run using exponential moving
/// averages over mean and stddev.
class AdaptiveNormalizer : public MovingStats {
public:
AdaptiveNormalizer(float smoothFactor=0.001f);
AdaptiveNormalizer(float mean, float stddev, float smoothFactor=0.001f);
virtual ~AdaptiveNormalizer() {}
void setMean(float mean) { _mean = mean; }
void setStddev(float stddev) { _stddev = stddev; };
virtual float put(float value);
virtual float get() { return _value; }
float _value;
float _mean;
float _stddev;
};
/// Standard normalizer: normalizes values on-the-run using real mean and stddev.
class Normalizer : public SimpleStats {
public:
Normalizer();
Normalizer(float mean, float stddev);
virtual ~Normalizer() {}
void setMean(float mean) { _mean = mean; }
void setStddev(float stddev) { _stddev = stddev; };
virtual float put(float value);
virtual float get() { return _value; }
float _value;
float _mean;
float _stddev;
};
/// Regularizes signal into [0,1] by rescaling it using the min and max values.
class MinMaxScaler {
public:
MinMaxScaler();
virtual ~MinMaxScaler() {}
virtual float put(float value);
virtual float get() { return _value; }
float _value;
float _minValue;
float _maxValue;
};
#endif // __CORESTATS_H__