oscura 0.0.1__py3-none-any.whl → 0.1.0__py3-none-any.whl

oscura/quality/ensemble.py

@@ -0,0 +1,740 @@
+"""Ensemble methods for combining multiple analysis algorithms.
+
+This module provides robust analysis by combining results from multiple algorithms
+using various aggregation strategies. Ensemble methods reduce individual algorithm
+bias, handle outliers, and provide confidence bounds for more reliable measurements.
+
+
+Example:
+    >>> from oscura.quality.ensemble import EnsembleAggregator, AggregationMethod
+    >>> from oscura.quality.ensemble import create_frequency_ensemble
+    >>> # Combine multiple frequency measurements
+    >>> result = create_frequency_ensemble(signal, sample_rate=1e9)
+    >>> print(f"Frequency: {result.value:.2f} Hz ± {result.confidence*100:.1f}%")
+    >>> print(f"Methods agree: {result.method_agreement*100:.1f}%")
+    >>> # Use custom ensemble
+    >>> aggregator = EnsembleAggregator(method=AggregationMethod.WEIGHTED_AVERAGE)
+    >>> results = [
+    ...     {"value": 1000.0, "confidence": 0.9, "method": "fft"},
+    ...     {"value": 1005.0, "confidence": 0.8, "method": "autocorr"},
+    ...     {"value": 995.0, "confidence": 0.85, "method": "zero_crossing"},
+    ... ]
+    >>> ensemble_result = aggregator.aggregate(results)
+
+References:
+    - Kuncheva, L.I.: "Combining Pattern Classifiers" (2nd Ed), Wiley, 2014
+    - Polikar, R.: "Ensemble Learning", Scholarpedia, 2009
+    - Dietterich, T.G.: "Ensemble Methods in Machine Learning", 2000
+"""
+
+from __future__ import annotations
+
+import logging
+from collections import Counter
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import TYPE_CHECKING, Any
+
+import numpy as np
+from scipy import stats
+
+from oscura.quality.scoring import AnalysisQualityScore, combine_quality_scores
+
+if TYPE_CHECKING:
+    from numpy.typing import NDArray
+
+logger = logging.getLogger(__name__)
+
+
+class AggregationMethod(Enum):
+    """Strategy for combining multiple analysis results.
+
+    Attributes:
+        WEIGHTED_AVERAGE: Weight results by confidence (best for numeric values).
+        VOTING: Majority voting (best for categorical results).
+        MEDIAN: Robust to outliers (best when outliers expected).
+        BAYESIAN: Bayesian combination with prior (best when prior knowledge available).
+    """
+
+    WEIGHTED_AVERAGE = "weighted_average"
+    VOTING = "voting"
+    MEDIAN = "median"
+    BAYESIAN = "bayesian"
+
+
+@dataclass
+class EnsembleResult:
+    """Combined result from multiple analysis methods.
+
+    Attributes:
+        value: Aggregated value (numeric or categorical).
+        confidence: Overall confidence in combined result (0-1).
+        lower_bound: Lower confidence bound (None for categorical).
+        upper_bound: Upper confidence bound (None for categorical).
+        method_agreement: Agreement between methods (0-1, higher is better).
+        individual_results: List of individual method results.
+        aggregation_method: Method used for aggregation.
+        quality_score: Optional quality score for the ensemble result.
+        outlier_methods: Indices of methods producing outlier results.
+
+    Example:
+        >>> if result.method_agreement > 0.8:
+        ...     print(f"High agreement: {result.value}")
+        >>> else:
+        ...     print(f"Methods disagree, confidence: {result.confidence}")
+    """
+
+    value: Any
+    confidence: float
+    lower_bound: float | None = None
+    upper_bound: float | None = None
+    method_agreement: float = 1.0
+    individual_results: list[dict[str, Any]] = field(default_factory=list)
+    aggregation_method: AggregationMethod = AggregationMethod.WEIGHTED_AVERAGE
+    quality_score: AnalysisQualityScore | None = None
+    outlier_methods: list[int] = field(default_factory=list)
+
+    def __post_init__(self) -> None:
+        """Validate confidence and agreement values."""
+        if not 0 <= self.confidence <= 1:
+            raise ValueError(f"Confidence must be in [0, 1], got {self.confidence}")
+        if not 0 <= self.method_agreement <= 1:
+            raise ValueError(f"Method agreement must be in [0, 1], got {self.method_agreement}")
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert to dictionary for serialization.
+
+        Returns:
+            Dictionary representation of ensemble result.
+        """
+        return {
+            "value": self.value,
+            "confidence": self.confidence,
+            "lower_bound": self.lower_bound,
+            "upper_bound": self.upper_bound,
+            "method_agreement": self.method_agreement,
+            "aggregation_method": self.aggregation_method.value,
+            "individual_results": self.individual_results,
+            "outlier_methods": self.outlier_methods,
+            "quality_score": self.quality_score.to_dict() if self.quality_score else None,
+        }
+
+
+class EnsembleAggregator:
+    """Combines multiple analysis results for robust estimation.
+
+    Supports various aggregation strategies optimized for different data types
+    and analysis scenarios. Automatically detects and handles outliers, computes
+    confidence bounds, and measures inter-method agreement.
+
+        QUAL-004: Ensemble Methods for Robust Analysis
+        QUAL-005: Disagreement Detection and Handling
+        QUAL-006: Confidence Bound Estimation
+
+    Example:
+        >>> aggregator = EnsembleAggregator(method=AggregationMethod.WEIGHTED_AVERAGE)
+        >>> results = [
+        ...     {"value": 100.0, "confidence": 0.9},
+        ...     {"value": 102.0, "confidence": 0.85},
+        ...     {"value": 98.0, "confidence": 0.8},
+        ... ]
+        >>> ensemble = aggregator.aggregate(results)
+        >>> print(f"Result: {ensemble.value:.2f} ± {ensemble.confidence*100:.1f}%")
+    """
+
+    def __init__(
+        self,
+        method: AggregationMethod = AggregationMethod.WEIGHTED_AVERAGE,
+        outlier_threshold: float = 3.0,
+        min_agreement: float = 0.5,
+    ):
+        """Initialize ensemble aggregator.
+
+        Args:
+            method: Aggregation strategy to use.
+            outlier_threshold: Z-score threshold for outlier detection (default 3.0).
+            min_agreement: Minimum agreement threshold to warn (default 0.5).
+        """
+        self.method = method
+        self.outlier_threshold = outlier_threshold
+        self.min_agreement = min_agreement
+
+    def aggregate(self, results: list[dict[str, Any]]) -> EnsembleResult:
+        """Combine multiple results into one robust estimate.
+
+        Args:
+            results: List of result dictionaries with keys:
+                - value: Measured value (numeric or categorical)
+                - confidence: Confidence score (0-1)
+                - method: Optional method name
+                - quality_score: Optional AnalysisQualityScore
+
+        Returns:
+            EnsembleResult with combined value and metadata.
+
+        Raises:
+            ValueError: If results list is empty or invalid.
+
+        Example:
+            >>> results = [
+            ...     {"value": 1000, "confidence": 0.9, "method": "fft"},
+            ...     {"value": 1005, "confidence": 0.85, "method": "autocorr"},
+            ... ]
+            >>> ensemble = aggregator.aggregate(results)
+        """
+        if not results:
+            raise ValueError("Cannot aggregate empty results list")
+
+        # Extract values and confidences
+        values = [r["value"] for r in results]
+        confidences = [r.get("confidence", 1.0) for r in results]
+
+        # Determine if values are numeric or categorical
+        is_numeric = all(isinstance(v, int | float | np.number) for v in values)
+
+        if is_numeric:
+            return self.aggregate_numeric(
+                [float(v) for v in values],
+                confidences,
+                original_results=results,
+            )
+        else:
+            return self.aggregate_categorical(
+                [str(v) for v in values],
+                confidences,
+                original_results=results,
+            )
+
+    def aggregate_numeric(
+        self,
+        values: list[float],
+        confidences: list[float],
+        original_results: list[dict[str, Any]] | None = None,
+    ) -> EnsembleResult:
+        """Combine numeric values with confidence weighting.
+
+        Args:
+            values: List of numeric values to combine.
+            confidences: Confidence scores for each value (0-1).
+            original_results: Optional original result dictionaries.
+
+        Returns:
+            EnsembleResult with aggregated numeric value.
+
+        Raises:
+            ValueError: If values list is empty.
+
+        Example:
+            >>> values = [100.0, 102.0, 98.0, 150.0]  # 150 is outlier
+            >>> confidences = [0.9, 0.85, 0.8, 0.7]
+            >>> result = aggregator.aggregate_numeric(values, confidences)
+            >>> # Outlier detected and handled
+        """
+        if not values:
+            raise ValueError("Cannot aggregate empty values list")
+
+        if original_results is None:
+            original_results = [
+                {"value": v, "confidence": c} for v, c in zip(values, confidences, strict=False)
+            ]
+
+        values_arr = np.array(values, dtype=np.float64)
+        confidences_arr = np.array(confidences, dtype=np.float64)
+
+        # Detect outliers
+        outlier_indices = self.detect_outlier_methods(original_results)
+
+        # Create mask for non-outlier values
+        valid_mask = np.ones(len(values), dtype=bool)
+        valid_mask[outlier_indices] = False
+
+        # Use only non-outliers for aggregation
+        valid_values = values_arr[valid_mask]
+        valid_confidences = confidences_arr[valid_mask]
+
+        if len(valid_values) == 0:
+            # All values are outliers, use all with warning
+            logger.warning("All methods detected as outliers, using all values")
+            valid_values = values_arr
+            valid_confidences = confidences_arr
+            outlier_indices = []
+
+        # Compute aggregated value based on method
+        if self.method == AggregationMethod.WEIGHTED_AVERAGE:
+            # Normalize weights
+            weights = valid_confidences / np.sum(valid_confidences)
+            aggregated_value = float(np.sum(valid_values * weights))
+            # Weighted variance
+            variance = float(np.sum(weights * (valid_values - aggregated_value) ** 2))
+            std_dev = np.sqrt(variance)
+
+        elif self.method == AggregationMethod.MEDIAN:
+            aggregated_value = float(np.median(valid_values))
+            # Use MAD (Median Absolute Deviation) for robust std estimate
+            mad = float(np.median(np.abs(valid_values - aggregated_value)))
+            std_dev = mad * 1.4826  # Scale factor for normal distribution
+
+        elif self.method == AggregationMethod.BAYESIAN:
+            # Bayesian combination with Gaussian likelihood
+            # Prior: uniform over range
+            # Likelihood: Gaussian with confidence-based variance
+            precisions = valid_confidences**2  # Higher confidence = lower variance
+            total_precision = np.sum(precisions)
+            aggregated_value = float(np.sum(valid_values * precisions) / total_precision)
+            # Posterior variance
+            variance = 1.0 / total_precision
+            std_dev = float(np.sqrt(variance))
+
+        else:
+            # Fallback to simple average
+            aggregated_value = float(np.mean(valid_values))
+            std_dev = float(np.std(valid_values))
+
+        # Compute confidence bounds (95% confidence interval)
+        if len(valid_values) > 1:
+            # Use t-distribution for small samples
+            dof = len(valid_values) - 1
+            t_value = stats.t.ppf(0.975, dof)  # 95% CI
+            margin = t_value * std_dev / np.sqrt(len(valid_values))
+            lower_bound = aggregated_value - margin
+            upper_bound = aggregated_value + margin
+        else:
+            lower_bound = aggregated_value
+            upper_bound = aggregated_value
+
+        # Compute method agreement (inverse of coefficient of variation)
+        if len(valid_values) > 1 and aggregated_value != 0:
+            cv = std_dev / abs(aggregated_value)
+            method_agreement = float(np.clip(1.0 - cv, 0.0, 1.0))
+        else:
+            method_agreement = 1.0
+
+        # Overall confidence (weighted average of individual confidences)
+        overall_confidence = float(np.mean(valid_confidences))
+
+        # Penalize confidence if agreement is low
+        if method_agreement < self.min_agreement:
+            overall_confidence *= method_agreement
+            logger.warning(
+                f"Low method agreement ({method_agreement:.2f}), "
+                f"reduced confidence to {overall_confidence:.2f}"
+            )
+
+        # Combine quality scores if available
+        quality_scores_raw = [
+            r.get("quality_score") for r in original_results if "quality_score" in r
+        ]
+        ensemble_quality = None
+        if quality_scores_raw and all(
+            isinstance(q, AnalysisQualityScore) for q in quality_scores_raw
+        ):
+            # Type narrowing - we know all are AnalysisQualityScore at this point
+            quality_scores: list[AnalysisQualityScore] = quality_scores_raw  # type: ignore[assignment]
+            ensemble_quality = combine_quality_scores(
+                quality_scores, weights=confidences[: len(quality_scores)]
+            )
+
+        return EnsembleResult(
+            value=aggregated_value,
+            confidence=overall_confidence,
+            lower_bound=lower_bound,
+            upper_bound=upper_bound,
+            method_agreement=method_agreement,
+            individual_results=original_results,
+            aggregation_method=self.method,
+            quality_score=ensemble_quality,
+            outlier_methods=outlier_indices,
+        )
+
+    def aggregate_categorical(
+        self,
+        values: list[str],
+        confidences: list[float],
+        original_results: list[dict[str, Any]] | None = None,
+    ) -> EnsembleResult:
+        """Combine categorical values via weighted voting.
+
+        Args:
+            values: List of categorical values to combine.
+            confidences: Confidence scores for each value (0-1).
+            original_results: Optional original result dictionaries.
+
+        Returns:
+            EnsembleResult with majority vote value.
+
+        Raises:
+            ValueError: If values list is empty.
+
+        Example:
+            >>> values = ["rising", "rising", "falling", "rising"]
+            >>> confidences = [0.9, 0.85, 0.6, 0.8]
+            >>> result = aggregator.aggregate_categorical(values, confidences)
+            >>> # "rising" wins by weighted vote
+        """
+        if not values:
+            raise ValueError("Cannot aggregate empty values list")
+
+        if original_results is None:
+            original_results = [
+                {"value": v, "confidence": c} for v, c in zip(values, confidences, strict=False)
+            ]
+
+        # Weighted voting
+        vote_weights: dict[str, float] = {}
+        for value, confidence in zip(values, confidences, strict=False):
+            vote_weights[value] = vote_weights.get(value, 0.0) + confidence
+
+        # Get winner
+        winner = max(vote_weights.items(), key=lambda x: x[1])
+        aggregated_value = winner[0]
+        total_weight = sum(vote_weights.values())
+
+        # Confidence is the fraction of votes for winner
+        overall_confidence = winner[1] / total_weight if total_weight > 0 else 0.0
+
+        # Agreement is measured by vote concentration
+        # Higher agreement when votes are concentrated on one option
+        vote_counts = Counter(values)
+        total_votes = len(values)
+        max_count = vote_counts.most_common(1)[0][1]
+        method_agreement = max_count / total_votes
+
+        # Combine quality scores if available
+        quality_scores_raw = [
+            r.get("quality_score") for r in original_results if "quality_score" in r
+        ]
+        ensemble_quality = None
+        if quality_scores_raw and all(
+            isinstance(q, AnalysisQualityScore) for q in quality_scores_raw
+        ):
+            # Type narrowing - we know all are AnalysisQualityScore at this point
+            quality_scores: list[AnalysisQualityScore] = quality_scores_raw  # type: ignore[assignment]
+            ensemble_quality = combine_quality_scores(
+                quality_scores, weights=confidences[: len(quality_scores)]
+            )
+
+        return EnsembleResult(
+            value=aggregated_value,
+            confidence=overall_confidence,
+            lower_bound=None,
+            upper_bound=None,
+            method_agreement=method_agreement,
+            individual_results=original_results,
+            aggregation_method=self.method,
+            quality_score=ensemble_quality,
+            outlier_methods=[],  # No outlier detection for categorical
+        )
+
+    def detect_outlier_methods(self, results: list[dict[str, Any]]) -> list[int]:
+        """Identify methods producing outlier results.
+
+        Uses modified Z-score (based on MAD) for robust outlier detection.
+
+        Args:
+            results: List of result dictionaries with "value" key.
+
+        Returns:
+            List of indices corresponding to outlier methods.
+
+        Example:
+            >>> results = [
+            ...     {"value": 100}, {"value": 102}, {"value": 98}, {"value": 500}
+            ... ]
+            >>> outliers = aggregator.detect_outlier_methods(results)
+            >>> # Returns [3] - the 500 value is an outlier
+        """
+        values = [r["value"] for r in results]
+
+        # Only works for numeric values
+        if not all(isinstance(v, int | float | np.number) for v in values):
+            return []
+
+        if len(values) < 3:
+            # Need at least 3 values for outlier detection
+            return []
+
+        values_arr = np.array(values, dtype=np.float64)
+
+        # Use modified Z-score based on MAD (robust to outliers)
+        median = np.median(values_arr)
+        mad = np.median(np.abs(values_arr - median))
+
+        if mad == 0:
+            # All values are identical
+            return []
+
+        # Modified Z-score
+        modified_z_scores = 0.6745 * (values_arr - median) / mad
+
+        # Identify outliers
+        outlier_mask = np.abs(modified_z_scores) > self.outlier_threshold
+        outlier_indices: list[int] = np.where(outlier_mask)[0].tolist()
+
+        if outlier_indices:
+            logger.info(f"Detected {len(outlier_indices)} outlier method(s): {outlier_indices}")
+
+        return outlier_indices
+
+
+# Pre-configured ensembles for common analysis types
+# Weights represent the relative reliability of each method
+
+FREQUENCY_ENSEMBLE: list[tuple[str, float]] = [
+    ("fft_peak", 0.4),  # FFT peak is generally most reliable
+    ("zero_crossing", 0.3),  # Zero crossing is robust but can be noisy
+    ("autocorrelation", 0.3),  # Autocorrelation handles noise well
+]
+
+EDGE_DETECTION_ENSEMBLE: list[tuple[str, float]] = [
+    ("threshold_crossing", 0.5),  # Most direct method
+    ("derivative", 0.3),  # Good for clean signals
+    ("schmitt_trigger", 0.2),  # Noise immunity but less precise
+]
+
+AMPLITUDE_ENSEMBLE: list[tuple[str, float]] = [
+    ("peak_to_peak", 0.4),  # Direct measurement
+    ("rms", 0.3),  # Robust to noise
+    ("percentile_99", 0.3),  # Outlier resistant
+]
+
+
+def create_frequency_ensemble(
+    signal: NDArray[np.float64],
+    sample_rate: float,
+    method_weights: list[tuple[str, float]] | None = None,
+) -> EnsembleResult:
+    """Run multiple frequency detection methods and combine results.
+
+    Applies FFT peak detection, zero-crossing rate, and autocorrelation-based
+    frequency estimation, then combines using weighted averaging.
+
+    Args:
+        signal: Input signal array.
+        sample_rate: Sample rate in Hz.
+        method_weights: Optional custom method weights. Defaults to FREQUENCY_ENSEMBLE.
+
+    Returns:
+        EnsembleResult with combined frequency estimate.
+
+    Raises:
+        ValueError: If all frequency detection methods fail.
+
+    Example:
+        >>> import numpy as np
+        >>> t = np.linspace(0, 1, 1000)
+        >>> signal = np.sin(2 * np.pi * 10 * t)  # 10 Hz sine
+        >>> result = create_frequency_ensemble(signal, sample_rate=1000)
+        >>> print(f"Frequency: {result.value:.2f} Hz")
+        >>> print(f"Confidence: {result.confidence:.2%}")
+    """
+    if method_weights is None:
+        method_weights = FREQUENCY_ENSEMBLE
+
+    results = []
+
+    # Method 1: FFT peak detection
+    try:
+        fft_result = np.fft.rfft(signal)
+        freqs = np.fft.rfftfreq(len(signal), d=1.0 / sample_rate)
+        peak_idx = np.argmax(np.abs(fft_result[1:])) + 1  # Skip DC
+        freq_fft = float(freqs[peak_idx])
+        # Confidence based on peak prominence
+        peak_magnitude = np.abs(fft_result[peak_idx])
+        mean_magnitude = np.mean(np.abs(fft_result[1:]))
+        confidence_fft = min(1.0, peak_magnitude / (mean_magnitude * 10))
+        results.append(
+            {
+                "value": freq_fft,
+                "confidence": confidence_fft * method_weights[0][1],
+                "method": "fft_peak",
+            }
+        )
+    except Exception as e:
+        logger.debug(f"FFT peak detection failed: {e}")
+
+    # Method 2: Zero crossing rate
+    try:
+        zero_crossings = np.where(np.diff(np.sign(signal)))[0]
+        if len(zero_crossings) > 1:
+            # Average time between zero crossings (half period)
+            avg_half_period = np.mean(np.diff(zero_crossings)) / sample_rate
+            freq_zc = 1.0 / (2.0 * avg_half_period)
+            # Confidence based on regularity of crossings
+            std_half_period = np.std(np.diff(zero_crossings)) / sample_rate
+            confidence_zc = max(0.0, 1.0 - std_half_period / avg_half_period)
+            results.append(
+                {
+                    "value": float(freq_zc),
+                    "confidence": confidence_zc * method_weights[1][1],
+                    "method": "zero_crossing",
+                }
+            )
+    except Exception as e:
+        logger.debug(f"Zero crossing detection failed: {e}")
+
+    # Method 3: Autocorrelation
+    try:
+        # Compute autocorrelation
+        autocorr = np.correlate(signal, signal, mode="full")
+        autocorr = autocorr[len(autocorr) // 2 :]
+        # Find first peak after zero lag (skip DC)
+        peaks = []
+        for i in range(1, min(len(autocorr) - 1, len(signal) // 2)):
+            if autocorr[i] > autocorr[i - 1] and autocorr[i] > autocorr[i + 1]:
+                peaks.append(i)
+        if peaks:
+            first_peak = peaks[0]
+            period_samples = first_peak
+            freq_ac = sample_rate / period_samples
+            # Confidence based on peak strength
+            peak_strength = autocorr[first_peak] / autocorr[0]
+            confidence_ac = float(np.clip(peak_strength, 0.0, 1.0))
+            results.append(
+                {
+                    "value": float(freq_ac),
+                    "confidence": confidence_ac * method_weights[2][1],
+                    "method": "autocorrelation",
+                }
+            )
+    except Exception as e:
+        logger.debug(f"Autocorrelation detection failed: {e}")
+
+    if not results:
+        raise ValueError("All frequency detection methods failed")
+
+    # Aggregate results
+    aggregator = EnsembleAggregator(method=AggregationMethod.WEIGHTED_AVERAGE)
+    return aggregator.aggregate(results)
+
+
+def create_edge_ensemble(
+    signal: NDArray[np.float64],
+    sample_rate: float,
+    threshold: float | None = None,
+    method_weights: list[tuple[str, float]] | None = None,
+) -> EnsembleResult:
+    """Run multiple edge detection methods and combine results.
+
+    Applies threshold crossing, derivative-based, and Schmitt trigger edge
+    detection, then combines results using weighted voting or averaging.
+
+    Args:
+        signal: Input signal array.
+        sample_rate: Sample rate in Hz.
+        threshold: Detection threshold. If None, uses signal midpoint.
+        method_weights: Optional custom method weights. Defaults to EDGE_DETECTION_ENSEMBLE.
+
+    Returns:
+        EnsembleResult with combined edge detection results.
+
+    Raises:
+        ValueError: If all edge detection methods fail.
+
+    Example:
+        >>> signal = np.array([0, 0, 1, 1, 0, 0, 1, 1])
+        >>> result = create_edge_ensemble(signal, sample_rate=1000)
+        >>> print(f"Edge count: {result.value}")
+        >>> print(f"Agreement: {result.method_agreement:.2%}")
+    """
+    if method_weights is None:
+        method_weights = EDGE_DETECTION_ENSEMBLE
+
+    if threshold is None:
+        threshold = float((np.max(signal) + np.min(signal)) / 2.0)
+
+    results = []
+
+    # Method 1: Threshold crossing
+    try:
+        crossings = np.where(np.diff(np.sign(signal - threshold)))[0]
+        edge_count_tc = len(crossings)
+        # Confidence based on signal quality (SNR proxy)
+        signal_range = np.ptp(signal)
+        noise_estimate = np.std(np.diff(signal))
+        confidence_tc = (
+            min(1.0, signal_range / (noise_estimate * 10)) if noise_estimate > 0 else 0.5
+        )
+        results.append(
+            {
+                "value": edge_count_tc,
+                "confidence": confidence_tc * method_weights[0][1],
+                "method": "threshold_crossing",
+            }
+        )
+    except Exception as e:
+        logger.debug(f"Threshold crossing detection failed: {e}")
+
+    # Method 2: Derivative-based
+    try:
+        derivative = np.diff(signal)
+        # Find peaks in absolute derivative
+        deriv_std = np.std(derivative)
+        deriv_threshold = deriv_std * 2
+        edge_indices = np.where(np.abs(derivative) > deriv_threshold)[0]
+        # Remove consecutive detections (within 2 samples)
+        filtered_edges = []
+        for i, idx in enumerate(edge_indices):
+            if i == 0 or idx - edge_indices[i - 1] > 2:
+                filtered_edges.append(idx)
+        edge_count_deriv = len(filtered_edges)
+        # Confidence based on peak derivative prominence above threshold
+        # Higher max derivative relative to threshold means clearer edges
+        max_deriv = np.max(np.abs(derivative)) if len(derivative) > 0 else 0.0
+        prominence_ratio = (max_deriv / deriv_threshold) if deriv_threshold > 0 else 0.0
+        confidence_deriv = float(
+            np.clip(prominence_ratio / 3.0, 0.0, 1.0)
+        )  # Normalize: 3x threshold = 100%
+        results.append(
+            {
+                "value": edge_count_deriv,
+                "confidence": confidence_deriv * method_weights[1][1],
+                "method": "derivative",
+            }
+        )
+    except Exception as e:
+        logger.debug(f"Derivative edge detection failed: {e}")
+
+    # Method 3: Schmitt trigger (hysteresis)
+    try:
+        hysteresis = float(np.std(signal) * 0.1)
+        thresh_high = threshold + hysteresis
+        thresh_low = threshold - hysteresis
+        state = signal[0] > threshold
+        edge_count_schmitt = 0
+        for val in signal:
+            if not state and val > thresh_high:
+                edge_count_schmitt += 1
+                state = True
+            elif state and val < thresh_low:
+                edge_count_schmitt += 1
+                state = False
+        # Confidence based on hysteresis effectiveness
+        confidence_schmitt = 0.7  # Lower base confidence due to hysteresis delay
+        results.append(
+            {
+                "value": edge_count_schmitt,
+                "confidence": confidence_schmitt * method_weights[2][1],
+                "method": "schmitt_trigger",
+            }
+        )
+    except Exception as e:
+        logger.debug(f"Schmitt trigger detection failed: {e}")
+
+    if not results:
+        raise ValueError("All edge detection methods failed")
+
+    # Aggregate results (use median for integer counts)
+    aggregator = EnsembleAggregator(method=AggregationMethod.MEDIAN)
+    return aggregator.aggregate(results)
+
+
+__all__ = [
+    "AMPLITUDE_ENSEMBLE",
+    "EDGE_DETECTION_ENSEMBLE",
+    "FREQUENCY_ENSEMBLE",
+    "AggregationMethod",
+    "EnsembleAggregator",
+    "EnsembleResult",
+    "create_edge_ensemble",
+    "create_frequency_ensemble",
+]