trend_explained_ratio()

double MayaFlux::Kinesis::Stochastic::Estimate::trend_explained_ratio ( std::span< const double >  samples )

staticnoexcept

Fraction of a span's variance attributable to its linear trend.

Parameters

samples

Values to analyze

Returns

1.0 when the span is well explained by a straight line from first to last sample, closer to 0.0 when the span's variance is dominated by fluctuation around that line rather than the trend itself

A cheap signal/noise split for a window: high trend-explained ratio means the window looks like real directed change, low means the window looks like jitter around a roughly fixed point.

Definition at line 270 of file Estimate.cpp.

{
    if (samples.size() < 2)
        return 0.0;
 
    const double first = samples.front();
    const double last = samples.back();
    const auto n = static_cast<double>(samples.size() - 1);
 
    const double total_mean = std::accumulate(samples.begin(), samples.end(), 0.0) / static_cast<double>(samples.size());
 
    double total_var = 0.0;
    double residual_var = 0.0;
 
    for (size_t i = 0; i < samples.size(); ++i) {
        const double t = static_cast<double>(i) / n;
        const double trend_value = first + t * (last - first);
 
        const double total_dev = samples[i] - total_mean;
        total_var += total_dev * total_dev;
 
        const double residual = samples[i] - trend_value;
        residual_var += residual * residual;
    }
 
    if (total_var < 1e-12)
        return 0.0;
 
    return std::clamp(1.0 - (residual_var / total_var), 0.0, 1.0);
}

Referenced by update_quiet_period(), and update_trend().

Here is the caller graph for this function: