accept()

template<typename T >

T MayaFlux::Kinesis::HampelFilter< T >::accept ( const T &  candidate )

inline

Test a candidate value and return the value to actually use.

Parameters

candidate

Newly computed value, typically straight out of a Kinesis::Differential function

Returns

candidate if it looks consistent with recent history, or the last accepted value if candidate looks like an isolated spike

The first call and the next (window - 1) calls always accept, since MAD is not meaningful against a window that has not yet filled with real data; a held value from an uninitialized filter would be an arbitrary default, not a real prior reading.

Definition at line 98 of file HampelFilter.hpp.

    {
        if (!m_has_history) {
            m_history.push(candidate);
            m_last_good = candidate;
            ++m_sample_count;
            if (m_sample_count >= m_history.capacity())
                m_has_history = true;
            return candidate;
        }
 
        const auto view = m_history.linearized_view();
 
        std::vector<double> magnitudes;
        magnitudes.reserve(view.size());
        for (const auto& v : view)
            magnitudes.push_back(magnitude_of(v));
 
        std::vector<double> sorted = magnitudes;
        std::ranges::sort(sorted);
        const double median_mag = sorted[sorted.size() / 2];
 
        std::vector<double> deviations;
        deviations.reserve(sorted.size());
        for (double m : sorted)
            deviations.push_back(std::abs(m - median_mag));
        std::ranges::sort(deviations);
        const double mad = deviations[deviations.size() / 2] * 1.4826;
 
        const double candidate_mag = magnitude_of(candidate);
        const bool looks_like_spike = (mad > 1e-12)
            && (std::abs(candidate_mag - median_mag) / mad > m_threshold_mad);
 
        if (looks_like_spike && m_consecutive_rejections < m_max_consecutive_rejections) {
            ++m_consecutive_rejections;
            return m_last_good;
        }
 
        if (looks_like_spike) {
            m_history.reset();
            m_sample_count = 0;
            m_has_history = false;
        }
 
        m_consecutive_rejections = 0;
        m_history.push(candidate);
        m_last_good = candidate;
        ++m_sample_count;
        if (m_sample_count >= m_history.capacity())
            m_has_history = true;
        return candidate;
    }

References MayaFlux::Kinesis::HampelFilter< T >::m_consecutive_rejections, MayaFlux::Kinesis::HampelFilter< T >::m_has_history, MayaFlux::Kinesis::HampelFilter< T >::m_history, MayaFlux::Kinesis::HampelFilter< T >::m_last_good, MayaFlux::Kinesis::HampelFilter< T >::m_max_consecutive_rejections, MayaFlux::Kinesis::HampelFilter< T >::m_sample_count, MayaFlux::Kinesis::HampelFilter< T >::m_threshold_mad, and MayaFlux::Kinesis::HampelFilter< T >::magnitude_of().

Here is the call graph for this function: