dspx 0.1.1-alpha.0

package/src/native/adapters/ZScoreNormalizeStage.h

@@ -0,0 +1,326 @@
+#pragma once
+
+#include "../IDspStage.h"
+#include "../core/MovingZScoreFilter.h"
+#include <vector>
+#include <string>
+#include <stdexcept>
+#include <cmath>
+#include <numeric>   // For std::accumulate
+#include <algorithm> // For std::max
+
+namespace dsp::adapters
+{
+    enum class ZScoreNormalizeMode
+    {
+        Batch,
+        Moving
+    };
+
+    class ZScoreNormalizeStage : public IDspStage
+    {
+    public:
+        /**
+         * @brief Constructs a new Z-Score Normalize Stage.
+         * @param mode The variance mode (Batch or Moving).
+         * @param window_size The window size in samples (0 if using duration-based).
+         * @param window_duration_ms The window duration in milliseconds (0 if using size-based).
+         * @param epsilon A small value to prevent division by zero (default 1e-6).
+         */
+        explicit ZScoreNormalizeStage(ZScoreNormalizeMode mode, size_t window_size = 0, double window_duration_ms = 0.0, float epsilon = 1e-6f)
+            : m_mode(mode),
+              m_window_size(window_size),
+              m_window_duration_ms(window_duration_ms),
+              m_epsilon(epsilon),
+              m_is_initialized(window_size > 0)
+        {
+            if (m_mode == ZScoreNormalizeMode::Moving && window_size == 0 && window_duration_ms == 0.0)
+            {
+                throw std::invalid_argument("ZScoreNormalize: either window size or window duration must be greater than 0 for 'moving' mode");
+            }
+        }
+
+        // Return the type identifier for this stage
+        const char *getType() const override
+        {
+            return "zScoreNormalize";
+        }
+
+        // Implementation of the interface method
+        void process(float *buffer, size_t numSamples, int numChannels, const float *timestamps = nullptr) override
+        {
+            if (m_mode == ZScoreNormalizeMode::Batch)
+            {
+                processBatch(buffer, numSamples, numChannels);
+            }
+            else // ZScoreNormalizeMode::Moving
+            {
+                processMoving(buffer, numSamples, numChannels, timestamps);
+            }
+        }
+
+        // Serialize the stage's state
+        Napi::Object serializeState(Napi::Env env) const override
+        {
+            Napi::Object state = Napi::Object::New(env);
+            std::string modeStr = (m_mode == ZScoreNormalizeMode::Moving) ? "moving" : "batch";
+            state.Set("mode", modeStr);
+            state.Set("epsilon", Napi::Number::New(env, m_epsilon));
+
+            if (m_mode == ZScoreNormalizeMode::Moving)
+            {
+                state.Set("windowSize", static_cast<uint32_t>(m_window_size));
+                state.Set("numChannels", static_cast<uint32_t>(m_filters.size()));
+
+                // Serialize each channel's filter state
+                Napi::Array channelsArray = Napi::Array::New(env, m_filters.size());
+                for (size_t i = 0; i < m_filters.size(); ++i)
+                {
+                    Napi::Object channelState = Napi::Object::New(env);
+
+                    // Get the filter's internal state
+                    auto [bufferData, sums] = m_filters[i].getState();
+                    auto [runningSum, runningSumOfSquares] = sums;
+
+                    // Convert buffer data to JavaScript array
+                    Napi::Array bufferArray = Napi::Array::New(env, bufferData.size());
+                    for (size_t j = 0; j < bufferData.size(); ++j)
+                    {
+                        bufferArray.Set(j, Napi::Number::New(env, bufferData[j]));
+                    }
+
+                    channelState.Set("buffer", bufferArray);
+                    channelState.Set("runningSum", Napi::Number::New(env, runningSum));
+                    channelState.Set("runningSumOfSquares", Napi::Number::New(env, runningSumOfSquares));
+
+                    channelsArray.Set(static_cast<uint32_t>(i), channelState);
+                }
+                state.Set("channels", channelsArray);
+            }
+
+            return state;
+        }
+
+        // Deserialize and restore the stage's state
+        void deserializeState(const Napi::Object &state) override
+        {
+            std::string modeStr = state.Get("mode").As<Napi::String>().Utf8Value();
+            ZScoreNormalizeMode newMode = (modeStr == "moving") ? ZScoreNormalizeMode::Moving : ZScoreNormalizeMode::Batch;
+
+            if (newMode != m_mode)
+            {
+                throw std::runtime_error("ZScoreNormalize mode mismatch during deserialization");
+            }
+
+            m_epsilon = state.Get("epsilon").As<Napi::Number>().FloatValue();
+
+            if (m_mode == ZScoreNormalizeMode::Moving)
+            {
+                // Get window size and validate
+                size_t windowSize = state.Get("windowSize").As<Napi::Number>().Uint32Value();
+                if (windowSize != m_window_size)
+                {
+                    throw std::runtime_error("Window size mismatch during deserialization");
+                }
+
+                // Get number of channels
+                uint32_t numChannels = state.Get("channels").As<Napi::Array>().Length();
+
+                // Recreate filters
+                m_filters.clear();
+                for (uint32_t i = 0; i < numChannels; ++i)
+                {
+                    m_filters.emplace_back(m_window_size, m_epsilon);
+                }
+
+                // Restore each channel's state (identical logic to VarianceStage)
+                Napi::Array channelsArray = state.Get("channels").As<Napi::Array>();
+                for (uint32_t i = 0; i < numChannels; ++i)
+                {
+                    Napi::Object channelState = channelsArray.Get(i).As<Napi::Object>();
+
+                    // Get buffer data
+                    Napi::Array bufferArray = channelState.Get("buffer").As<Napi::Array>();
+                    std::vector<float> bufferData;
+                    bufferData.reserve(bufferArray.Length());
+                    for (uint32_t j = 0; j < bufferArray.Length(); ++j)
+                    {
+                        bufferData.push_back(bufferArray.Get(j).As<Napi::Number>().FloatValue());
+                    }
+
+                    // Get running sums
+                    float runningSum = channelState.Get("runningSum").As<Napi::Number>().FloatValue();
+                    float runningSumOfSquares = channelState.Get("runningSumOfSquares").As<Napi::Number>().FloatValue();
+
+                    // --- Validation (identical to VarianceStage) ---
+                    float actualSum = 0.0f;
+                    float actualSumOfSquares = 0.0f;
+                    for (const auto &val : bufferData)
+                    {
+                        actualSum += val;
+                        actualSumOfSquares += val * val;
+                    }
+
+                    const float toleranceSum = 0.0001f * std::max(1.0f, std::abs(actualSum));
+                    if (std::abs(runningSum - actualSum) > toleranceSum)
+                    {
+                        throw std::runtime_error(
+                            "Running sum validation failed: expected " +
+                            std::to_string(actualSum) + " but got " +
+                            std::to_string(runningSum));
+                    }
+
+                    const float toleranceSq = 0.0001f * std::max(1.0f, std::abs(actualSumOfSquares));
+                    if (std::abs(runningSumOfSquares - actualSumOfSquares) > toleranceSq)
+                    {
+                        throw std::runtime_error(
+                            "Running sum of squares validation failed: expected " +
+                            std::to_string(actualSumOfSquares) + " but got " +
+                            std::to_string(runningSumOfSquares));
+                    }
+                    // --- End Validation ---
+
+                    // Restore the filter's state
+                    m_filters[i].setState(bufferData, runningSum, runningSumOfSquares);
+                }
+            }
+        }
+
+        // Reset all filters to initial state
+        void reset() override
+        {
+            for (auto &filter : m_filters)
+            {
+                filter.clear();
+            }
+        }
+
+    private:
+        /**
+         * @brief Statelessly calculates the Z-Score for each sample
+         * based on the entire buffer's stats for that channel.
+         */
+        void processBatch(float *buffer, size_t numSamples, int numChannels)
+        {
+            for (int c = 0; c < numChannels; ++c)
+            {
+                size_t numSamplesPerChannel = numSamples / numChannels;
+                if (numSamplesPerChannel == 0)
+                    continue;
+
+                double sum = 0.0;
+                double sum_sq = 0.0;
+
+                // First pass: Calculate sums
+                for (size_t i = c; i < numSamples; i += numChannels)
+                {
+                    double val = static_cast<double>(buffer[i]);
+                    sum += val;
+                    sum_sq += val * val;
+                }
+
+                // Calculate mean and variance
+                double mean = sum / numSamplesPerChannel;
+                double mean_sq = sum_sq / numSamplesPerChannel;
+                double variance = std::max(0.0, mean_sq - (mean * mean));
+                float stddev = static_cast<float>(std::sqrt(variance));
+
+                // Second pass: Apply Z-Score normalization in-place
+                if (stddev < m_epsilon)
+                {
+                    // StdDev is zero, all values are the mean, Z-Score is 0
+                    for (size_t i = c; i < numSamples; i += numChannels)
+                    {
+                        buffer[i] = 0.0f;
+                    }
+                }
+                else
+                {
+                    float mean_f = static_cast<float>(mean);
+                    for (size_t i = c; i < numSamples; i += numChannels)
+                    {
+                        buffer[i] = (buffer[i] - mean_f) / stddev;
+                    }
+                }
+            }
+        }
+
+        /**
+         * @brief Statefully processes samples using the moving z-score filters.
+         */
+        void processMoving(float *buffer, size_t numSamples, int numChannels, const float *timestamps)
+        {
+            // Determine if we're in time-aware mode
+            bool useTimeAware = (m_window_duration_ms > 0.0) && timestamps != nullptr;
+
+            // Lazy initialization: convert windowDuration to windowSize if needed
+            if (!m_is_initialized && m_window_duration_ms > 0.0)
+            {
+                if (timestamps != nullptr && numSamples > 1)
+                {
+                    // Estimate sample rate from timestamps
+                    size_t samples_to_check = std::min(numSamples, size_t(10));
+                    double total_time_ms = timestamps[samples_to_check - 1] - timestamps[0];
+                    double avg_sample_period_ms = total_time_ms / (samples_to_check - 1);
+                    double estimated_sample_rate = 1000.0 / avg_sample_period_ms;
+
+                    // Use 3x the estimated size for time-aware mode
+                    size_t estimated_size = static_cast<size_t>((m_window_duration_ms / 1000.0) * estimated_sample_rate);
+                    m_window_size = std::max(size_t(1), estimated_size * 3);
+
+                    m_is_initialized = true;
+                }
+                else
+                {
+                    throw std::runtime_error("ZScoreNormalize: windowDuration was set, but timestamps are not available to derive sample rate");
+                }
+            }
+
+            // Lazily initialize our filters, one for each channel
+            if (m_filters.size() != static_cast<size_t>(numChannels))
+            {
+                m_filters.clear();
+                for (int i = 0; i < numChannels; ++i)
+                {
+                    if (useTimeAware)
+                    {
+                        // Create time-aware filter
+                        m_filters.emplace_back(m_window_size, m_window_duration_ms, m_epsilon);
+                    }
+                    else
+                    {
+                        // Create regular filter
+                        m_filters.emplace_back(m_window_size, m_epsilon);
+                    }
+                }
+            }
+
+            // Process the buffer sample by sample, de-interleaving
+            for (size_t i = 0; i < numSamples; ++i)
+            {
+                int channel = i % numChannels;
+                size_t sample_index = i / numChannels;
+
+                if (useTimeAware)
+                {
+                    // Use time-aware processing
+                    buffer[i] = m_filters[channel].addSampleWithTimestamp(buffer[i], timestamps[sample_index]);
+                }
+                else
+                {
+                    // Use sample-count processing
+                    buffer[i] = m_filters[channel].addSample(buffer[i]);
+                }
+            }
+        }
+
+        ZScoreNormalizeMode m_mode;
+        size_t m_window_size;
+        double m_window_duration_ms;
+        float m_epsilon;
+        bool m_is_initialized;
+        // We need a separate filter instance for each channel's state
+        std::vector<dsp::core::MovingZScoreFilter<float>> m_filters;
+    };
+
+} // namespace dsp::adapters