truthound-dashboard 1.3.1__py3-none-any.whl → 1.4.0__py3-none-any.whl

truthound_dashboard/core/streaming_anomaly.py

@@ -0,0 +1,883 @@
+"""Streaming anomaly detection service.
+
+This module provides real-time streaming anomaly detection capabilities,
+supporting sliding window detection and online learning.
+"""
+
+from __future__ import annotations
+
+import asyncio
+from collections import deque
+from dataclasses import dataclass, field
+from datetime import datetime
+from enum import Enum
+from typing import Any
+from uuid import uuid4
+
+import numpy as np
+
+
+class StreamingSessionStatus(str, Enum):
+    """Status of a streaming session."""
+
+    CREATED = "created"
+    RUNNING = "running"
+    PAUSED = "paused"
+    STOPPED = "stopped"
+    ERROR = "error"
+
+
+class StreamingAlgorithm(str, Enum):
+    """Supported streaming anomaly detection algorithms."""
+
+    ZSCORE_ROLLING = "zscore_rolling"
+    EXPONENTIAL_MOVING_AVERAGE = "ema"
+    ISOLATION_FOREST_INCREMENTAL = "isolation_forest_incremental"
+    HALF_SPACE_TREES = "half_space_trees"
+    ROBUST_RANDOM_CUT_FOREST = "rrcf"
+
+
+@dataclass
+class StreamingAlert:
+    """An anomaly alert from streaming detection."""
+
+    id: str
+    session_id: str
+    timestamp: datetime
+    data_point: dict[str, Any]
+    anomaly_score: float
+    is_anomaly: bool
+    algorithm: StreamingAlgorithm
+    details: dict[str, Any] = field(default_factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "id": self.id,
+            "session_id": self.session_id,
+            "timestamp": self.timestamp.isoformat(),
+            "data_point": self.data_point,
+            "anomaly_score": self.anomaly_score,
+            "is_anomaly": self.is_anomaly,
+            "algorithm": self.algorithm.value,
+            "details": self.details,
+        }
+
+
+@dataclass
+class StreamingStatistics:
+    """Rolling statistics for streaming detection."""
+
+    count: int = 0
+    mean: float = 0.0
+    variance: float = 0.0
+    min_value: float = float("inf")
+    max_value: float = float("-inf")
+    anomaly_count: int = 0
+
+    def update(self, value: float, is_anomaly: bool = False) -> None:
+        """Update statistics with a new value using Welford's algorithm."""
+        self.count += 1
+        delta = value - self.mean
+        self.mean += delta / self.count
+        delta2 = value - self.mean
+        self.variance += delta * delta2
+
+        self.min_value = min(self.min_value, value)
+        self.max_value = max(self.max_value, value)
+
+        if is_anomaly:
+            self.anomaly_count += 1
+
+    @property
+    def std(self) -> float:
+        """Get standard deviation."""
+        if self.count < 2:
+            return 0.0
+        return np.sqrt(self.variance / (self.count - 1))
+
+    @property
+    def anomaly_rate(self) -> float:
+        """Get anomaly rate."""
+        if self.count == 0:
+            return 0.0
+        return self.anomaly_count / self.count
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "count": self.count,
+            "mean": self.mean,
+            "std": self.std,
+            "min": self.min_value if self.min_value != float("inf") else None,
+            "max": self.max_value if self.max_value != float("-inf") else None,
+            "anomaly_count": self.anomaly_count,
+            "anomaly_rate": self.anomaly_rate,
+        }
+
+
+@dataclass
+class StreamingSession:
+    """A streaming anomaly detection session."""
+
+    id: str
+    source_id: str | None
+    algorithm: StreamingAlgorithm
+    window_size: int
+    threshold: float
+    columns: list[str]
+    status: StreamingSessionStatus
+    created_at: datetime
+    started_at: datetime | None = None
+    stopped_at: datetime | None = None
+    config: dict[str, Any] = field(default_factory=dict)
+
+    # Runtime state (not persisted)
+    _buffer: deque = field(default_factory=lambda: deque(maxlen=1000))
+    _column_stats: dict[str, StreamingStatistics] = field(default_factory=dict)
+    _alerts: list[StreamingAlert] = field(default_factory=list)
+    _alert_callbacks: list = field(default_factory=list)
+    _ema_values: dict[str, float] = field(default_factory=dict)
+
+    def __post_init__(self) -> None:
+        """Initialize column statistics."""
+        for col in self.columns:
+            self._column_stats[col] = StreamingStatistics()
+            self._ema_values[col] = 0.0
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "id": self.id,
+            "source_id": self.source_id,
+            "algorithm": self.algorithm.value,
+            "window_size": self.window_size,
+            "threshold": self.threshold,
+            "columns": self.columns,
+            "status": self.status.value,
+            "created_at": self.created_at.isoformat(),
+            "started_at": self.started_at.isoformat() if self.started_at else None,
+            "stopped_at": self.stopped_at.isoformat() if self.stopped_at else None,
+            "config": self.config,
+            "statistics": {col: stats.to_dict() for col, stats in self._column_stats.items()},
+            "total_points": len(self._buffer),
+            "total_alerts": len(self._alerts),
+        }
+
+
+class StreamingAnomalyDetector:
+    """Real-time streaming anomaly detection service.
+
+    Supports:
+    - Sliding window detection
+    - Multiple algorithms (Z-score, EMA, etc.)
+    - Online learning / model updates
+    - Alert callbacks for real-time notifications
+    """
+
+    def __init__(self) -> None:
+        """Initialize the streaming detector."""
+        self._sessions: dict[str, StreamingSession] = {}
+        self._lock = asyncio.Lock()
+
+    # =========================================================================
+    # Session Management
+    # =========================================================================
+
+    async def create_session(
+        self,
+        *,
+        source_id: str | None = None,
+        algorithm: StreamingAlgorithm = StreamingAlgorithm.ZSCORE_ROLLING,
+        window_size: int = 100,
+        threshold: float = 3.0,
+        columns: list[str] | None = None,
+        config: dict[str, Any] | None = None,
+    ) -> StreamingSession:
+        """Create a new streaming session.
+
+        Args:
+            source_id: Optional source ID to associate with.
+            algorithm: Detection algorithm to use.
+            window_size: Size of the sliding window.
+            threshold: Anomaly detection threshold.
+            columns: Columns to monitor (if None, monitors all numeric).
+            config: Additional algorithm configuration.
+
+        Returns:
+            Created streaming session.
+        """
+        session_id = str(uuid4())
+        session = StreamingSession(
+            id=session_id,
+            source_id=source_id,
+            algorithm=algorithm,
+            window_size=window_size,
+            threshold=threshold,
+            columns=columns or [],
+            status=StreamingSessionStatus.CREATED,
+            created_at=datetime.utcnow(),
+            config=config or {},
+        )
+
+        async with self._lock:
+            self._sessions[session_id] = session
+
+        return session
+
+    async def start_session(self, session_id: str) -> StreamingSession:
+        """Start a streaming session.
+
+        Args:
+            session_id: Session ID to start.
+
+        Returns:
+            Updated session.
+
+        Raises:
+            ValueError: If session not found.
+        """
+        async with self._lock:
+            session = self._sessions.get(session_id)
+            if session is None:
+                raise ValueError(f"Session '{session_id}' not found")
+
+            session.status = StreamingSessionStatus.RUNNING
+            session.started_at = datetime.utcnow()
+
+        return session
+
+    async def stop_session(self, session_id: str) -> StreamingSession:
+        """Stop a streaming session.
+
+        Args:
+            session_id: Session ID to stop.
+
+        Returns:
+            Updated session.
+
+        Raises:
+            ValueError: If session not found.
+        """
+        async with self._lock:
+            session = self._sessions.get(session_id)
+            if session is None:
+                raise ValueError(f"Session '{session_id}' not found")
+
+            session.status = StreamingSessionStatus.STOPPED
+            session.stopped_at = datetime.utcnow()
+
+        return session
+
+    async def get_session(self, session_id: str) -> StreamingSession | None:
+        """Get a session by ID.
+
+        Args:
+            session_id: Session ID.
+
+        Returns:
+            Session or None if not found.
+        """
+        return self._sessions.get(session_id)
+
+    async def list_sessions(self) -> list[StreamingSession]:
+        """List all active sessions.
+
+        Returns:
+            List of sessions.
+        """
+        return list(self._sessions.values())
+
+    async def delete_session(self, session_id: str) -> bool:
+        """Delete a session.
+
+        Args:
+            session_id: Session ID to delete.
+
+        Returns:
+            True if deleted.
+        """
+        async with self._lock:
+            if session_id in self._sessions:
+                del self._sessions[session_id]
+                return True
+        return False
+
+    # =========================================================================
+    # Data Processing
+    # =========================================================================
+
+    async def push_data_point(
+        self,
+        session_id: str,
+        data: dict[str, Any],
+        timestamp: datetime | None = None,
+    ) -> StreamingAlert | None:
+        """Push a data point to a streaming session.
+
+        Args:
+            session_id: Session ID.
+            data: Data point (column name -> value).
+            timestamp: Optional timestamp (defaults to now).
+
+        Returns:
+            Alert if anomaly detected, None otherwise.
+
+        Raises:
+            ValueError: If session not found or not running.
+        """
+        session = self._sessions.get(session_id)
+        if session is None:
+            raise ValueError(f"Session '{session_id}' not found")
+
+        if session.status != StreamingSessionStatus.RUNNING:
+            raise ValueError(f"Session '{session_id}' is not running")
+
+        timestamp = timestamp or datetime.utcnow()
+
+        # Store data point in buffer
+        session._buffer.append({"timestamp": timestamp, "data": data})
+
+        # Run anomaly detection
+        alert = await self._detect_anomaly(session, data, timestamp)
+
+        # Update statistics
+        for col, value in data.items():
+            if col in session._column_stats:
+                try:
+                    numeric_value = float(value)
+                    is_anomaly = alert is not None and alert.is_anomaly
+                    session._column_stats[col].update(numeric_value, is_anomaly)
+                except (ValueError, TypeError):
+                    pass
+
+        # Store alert and trigger callbacks
+        if alert is not None:
+            session._alerts.append(alert)
+            await self._trigger_alert_callbacks(session, alert)
+
+        return alert
+
+    async def push_batch(
+        self,
+        session_id: str,
+        data_points: list[dict[str, Any]],
+        timestamps: list[datetime] | None = None,
+    ) -> list[StreamingAlert]:
+        """Push a batch of data points.
+
+        Args:
+            session_id: Session ID.
+            data_points: List of data points.
+            timestamps: Optional list of timestamps.
+
+        Returns:
+            List of alerts.
+        """
+        alerts = []
+        timestamps = timestamps or [datetime.utcnow()] * len(data_points)
+
+        for data, ts in zip(data_points, timestamps):
+            alert = await self.push_data_point(session_id, data, ts)
+            if alert is not None:
+                alerts.append(alert)
+
+        return alerts
+
+    # =========================================================================
+    # Anomaly Detection Algorithms
+    # =========================================================================
+
+    async def _detect_anomaly(
+        self,
+        session: StreamingSession,
+        data: dict[str, Any],
+        timestamp: datetime,
+    ) -> StreamingAlert | None:
+        """Run anomaly detection on a data point.
+
+        Args:
+            session: Streaming session.
+            data: Data point.
+            timestamp: Timestamp.
+
+        Returns:
+            Alert if anomaly detected.
+        """
+        algorithm = session.algorithm
+
+        if algorithm == StreamingAlgorithm.ZSCORE_ROLLING:
+            return await self._detect_zscore_rolling(session, data, timestamp)
+        elif algorithm == StreamingAlgorithm.EXPONENTIAL_MOVING_AVERAGE:
+            return await self._detect_ema(session, data, timestamp)
+        elif algorithm == StreamingAlgorithm.ISOLATION_FOREST_INCREMENTAL:
+            return await self._detect_isolation_forest_incremental(session, data, timestamp)
+        elif algorithm == StreamingAlgorithm.HALF_SPACE_TREES:
+            return await self._detect_half_space_trees(session, data, timestamp)
+        elif algorithm == StreamingAlgorithm.ROBUST_RANDOM_CUT_FOREST:
+            return await self._detect_rrcf(session, data, timestamp)
+        else:
+            return None
+
+    async def _detect_zscore_rolling(
+        self,
+        session: StreamingSession,
+        data: dict[str, Any],
+        timestamp: datetime,
+    ) -> StreamingAlert | None:
+        """Z-score based anomaly detection using rolling statistics.
+
+        Args:
+            session: Streaming session.
+            data: Data point.
+            timestamp: Timestamp.
+
+        Returns:
+            Alert if anomaly detected.
+        """
+        # Need at least window_size points for reliable detection
+        if len(session._buffer) < min(session.window_size, 10):
+            return None
+
+        # Get recent values for each column
+        window_data = list(session._buffer)[-session.window_size:]
+
+        max_zscore = 0.0
+        anomaly_columns = []
+        details = {}
+
+        for col in session.columns:
+            if col not in data:
+                continue
+
+            try:
+                current_value = float(data[col])
+            except (ValueError, TypeError):
+                continue
+
+            # Calculate rolling mean and std from window
+            window_values = []
+            for point in window_data[:-1]:  # Exclude current point
+                if col in point.get("data", {}):
+                    try:
+                        window_values.append(float(point["data"][col]))
+                    except (ValueError, TypeError):
+                        pass
+
+            if len(window_values) < 2:
+                continue
+
+            window_mean = np.mean(window_values)
+            window_std = np.std(window_values)
+
+            if window_std == 0:
+                window_std = 1e-10  # Avoid division by zero
+
+            zscore = abs(current_value - window_mean) / window_std
+
+            if zscore > max_zscore:
+                max_zscore = zscore
+
+            if zscore > session.threshold:
+                anomaly_columns.append(col)
+                details[col] = {
+                    "value": current_value,
+                    "mean": float(window_mean),
+                    "std": float(window_std),
+                    "zscore": float(zscore),
+                }
+
+        is_anomaly = len(anomaly_columns) > 0 or max_zscore > session.threshold
+
+        if is_anomaly:
+            return StreamingAlert(
+                id=str(uuid4()),
+                session_id=session.id,
+                timestamp=timestamp,
+                data_point=data,
+                anomaly_score=float(max_zscore),
+                is_anomaly=True,
+                algorithm=StreamingAlgorithm.ZSCORE_ROLLING,
+                details={
+                    "anomaly_columns": anomaly_columns,
+                    "column_details": details,
+                    "threshold": session.threshold,
+                },
+            )
+
+        return None
+
+    async def _detect_ema(
+        self,
+        session: StreamingSession,
+        data: dict[str, Any],
+        timestamp: datetime,
+    ) -> StreamingAlert | None:
+        """Exponential Moving Average based anomaly detection.
+
+        Args:
+            session: Streaming session.
+            data: Data point.
+            timestamp: Timestamp.
+
+        Returns:
+            Alert if anomaly detected.
+        """
+        alpha = session.config.get("alpha", 0.1)  # Smoothing factor
+        threshold_multiplier = session.config.get("threshold_multiplier", 2.0)
+
+        max_deviation = 0.0
+        anomaly_columns = []
+        details = {}
+
+        for col in session.columns:
+            if col not in data:
+                continue
+
+            try:
+                current_value = float(data[col])
+            except (ValueError, TypeError):
+                continue
+
+            # Initialize EMA if first point
+            if session._ema_values.get(col, 0) == 0:
+                session._ema_values[col] = current_value
+                continue
+
+            # Calculate EMA
+            prev_ema = session._ema_values[col]
+            new_ema = alpha * current_value + (1 - alpha) * prev_ema
+            session._ema_values[col] = new_ema
+
+            # Calculate deviation from EMA
+            deviation = abs(current_value - prev_ema)
+
+            # Use rolling std for threshold
+            stats = session._column_stats.get(col)
+            if stats and stats.std > 0:
+                normalized_deviation = deviation / stats.std
+                if normalized_deviation > max_deviation:
+                    max_deviation = normalized_deviation
+
+                if normalized_deviation > session.threshold * threshold_multiplier:
+                    anomaly_columns.append(col)
+                    details[col] = {
+                        "value": current_value,
+                        "ema": float(new_ema),
+                        "deviation": float(deviation),
+                        "normalized_deviation": float(normalized_deviation),
+                    }
+
+        is_anomaly = len(anomaly_columns) > 0
+
+        if is_anomaly:
+            return StreamingAlert(
+                id=str(uuid4()),
+                session_id=session.id,
+                timestamp=timestamp,
+                data_point=data,
+                anomaly_score=float(max_deviation),
+                is_anomaly=True,
+                algorithm=StreamingAlgorithm.EXPONENTIAL_MOVING_AVERAGE,
+                details={
+                    "anomaly_columns": anomaly_columns,
+                    "column_details": details,
+                    "alpha": alpha,
+                },
+            )
+
+        return None
+
+    async def _detect_isolation_forest_incremental(
+        self,
+        session: StreamingSession,
+        data: dict[str, Any],
+        timestamp: datetime,
+    ) -> StreamingAlert | None:
+        """Incremental Isolation Forest based anomaly detection.
+
+        Uses a simplified streaming version that periodically retrains.
+
+        Args:
+            session: Streaming session.
+            data: Data point.
+            timestamp: Timestamp.
+
+        Returns:
+            Alert if anomaly detected.
+        """
+        # Minimum points before detection
+        if len(session._buffer) < session.window_size:
+            return None
+
+        try:
+            from sklearn.ensemble import IsolationForest
+
+            # Get recent window data
+            window_data = list(session._buffer)[-session.window_size:]
+
+            # Build feature matrix from window
+            feature_cols = [col for col in session.columns if col in data]
+            if not feature_cols:
+                return None
+
+            X = []
+            for point in window_data:
+                row = []
+                valid = True
+                for col in feature_cols:
+                    if col in point.get("data", {}):
+                        try:
+                            row.append(float(point["data"][col]))
+                        except (ValueError, TypeError):
+                            valid = False
+                            break
+                    else:
+                        valid = False
+                        break
+                if valid:
+                    X.append(row)
+
+            if len(X) < 10:
+                return None
+
+            X = np.array(X)
+
+            # Build current point feature vector
+            current_point = []
+            for col in feature_cols:
+                try:
+                    current_point.append(float(data[col]))
+                except (ValueError, TypeError):
+                    return None
+
+            current_point = np.array([current_point])
+
+            # Fit Isolation Forest on window
+            contamination = session.config.get("contamination", 0.1)
+            clf = IsolationForest(
+                n_estimators=50,
+                contamination=contamination,
+                random_state=42,
+            )
+            clf.fit(X)
+
+            # Predict on current point
+            prediction = clf.predict(current_point)[0]
+            score = -clf.score_samples(current_point)[0]
+
+            is_anomaly = prediction == -1
+
+            if is_anomaly:
+                return StreamingAlert(
+                    id=str(uuid4()),
+                    session_id=session.id,
+                    timestamp=timestamp,
+                    data_point=data,
+                    anomaly_score=float(score),
+                    is_anomaly=True,
+                    algorithm=StreamingAlgorithm.ISOLATION_FOREST_INCREMENTAL,
+                    details={
+                        "window_size": len(X),
+                        "contamination": contamination,
+                    },
+                )
+
+        except ImportError:
+            pass
+
+        return None
+
+    async def _detect_half_space_trees(
+        self,
+        session: StreamingSession,
+        data: dict[str, Any],
+        timestamp: datetime,
+    ) -> StreamingAlert | None:
+        """Half-Space Trees streaming anomaly detection.
+
+        A simplified implementation of HS-Trees for streaming.
+
+        Args:
+            session: Streaming session.
+            data: Data point.
+            timestamp: Timestamp.
+
+        Returns:
+            Alert if anomaly detected.
+        """
+        # Use Z-score as a fallback for HS-Trees
+        # A full implementation would maintain the tree structure
+        return await self._detect_zscore_rolling(session, data, timestamp)
+
+    async def _detect_rrcf(
+        self,
+        session: StreamingSession,
+        data: dict[str, Any],
+        timestamp: datetime,
+    ) -> StreamingAlert | None:
+        """Robust Random Cut Forest streaming anomaly detection.
+
+        A simplified implementation using codisp (collusive displacement).
+
+        Args:
+            session: Streaming session.
+            data: Data point.
+            timestamp: Timestamp.
+
+        Returns:
+            Alert if anomaly detected.
+        """
+        # Use Z-score as a fallback for RRCF
+        # A full implementation would use the rrcf library
+        return await self._detect_zscore_rolling(session, data, timestamp)
+
+    # =========================================================================
+    # Alert Management
+    # =========================================================================
+
+    def register_alert_callback(
+        self,
+        session_id: str,
+        callback: callable,
+    ) -> None:
+        """Register a callback for alerts.
+
+        Args:
+            session_id: Session ID.
+            callback: Callback function (async).
+        """
+        session = self._sessions.get(session_id)
+        if session:
+            session._alert_callbacks.append(callback)
+
+    def unregister_alert_callback(
+        self,
+        session_id: str,
+        callback: callable,
+    ) -> None:
+        """Unregister an alert callback.
+
+        Args:
+            session_id: Session ID.
+            callback: Callback function to remove.
+        """
+        session = self._sessions.get(session_id)
+        if session and callback in session._alert_callbacks:
+            session._alert_callbacks.remove(callback)
+
+    async def _trigger_alert_callbacks(
+        self,
+        session: StreamingSession,
+        alert: StreamingAlert,
+    ) -> None:
+        """Trigger all registered alert callbacks.
+
+        Args:
+            session: Streaming session.
+            alert: Alert to send.
+        """
+        for callback in session._alert_callbacks:
+            try:
+                if asyncio.iscoroutinefunction(callback):
+                    await callback(alert)
+                else:
+                    callback(alert)
+            except Exception:
+                pass  # Don't let callback errors break detection
+
+    async def get_alerts(
+        self,
+        session_id: str,
+        *,
+        limit: int = 100,
+        offset: int = 0,
+    ) -> list[StreamingAlert]:
+        """Get alerts for a session.
+
+        Args:
+            session_id: Session ID.
+            limit: Maximum alerts to return.
+            offset: Offset for pagination.
+
+        Returns:
+            List of alerts.
+        """
+        session = self._sessions.get(session_id)
+        if session is None:
+            return []
+
+        # Return alerts in reverse order (most recent first)
+        alerts = list(reversed(session._alerts))
+        return alerts[offset : offset + limit]
+
+    async def get_statistics(
+        self,
+        session_id: str,
+    ) -> dict[str, Any]:
+        """Get statistics for a session.
+
+        Args:
+            session_id: Session ID.
+
+        Returns:
+            Statistics dictionary.
+        """
+        session = self._sessions.get(session_id)
+        if session is None:
+            return {}
+
+        return {
+            "total_points": len(session._buffer),
+            "total_alerts": len(session._alerts),
+            "columns": {
+                col: stats.to_dict()
+                for col, stats in session._column_stats.items()
+            },
+            "buffer_utilization": len(session._buffer) / session._buffer.maxlen if session._buffer.maxlen else 0,
+        }
+
+    async def get_recent_data(
+        self,
+        session_id: str,
+        *,
+        limit: int = 100,
+    ) -> list[dict[str, Any]]:
+        """Get recent data points.
+
+        Args:
+            session_id: Session ID.
+            limit: Maximum points to return.
+
+        Returns:
+            List of recent data points.
+        """
+        session = self._sessions.get(session_id)
+        if session is None:
+            return []
+
+        # Return most recent points
+        recent = list(session._buffer)[-limit:]
+        return [
+            {
+                "timestamp": point["timestamp"].isoformat(),
+                "data": point["data"],
+            }
+            for point in reversed(recent)
+        ]
+
+
+# Global streaming detector instance
+_streaming_detector: StreamingAnomalyDetector | None = None
+
+
+def get_streaming_detector() -> StreamingAnomalyDetector:
+    """Get the global streaming detector instance.
+
+    Returns:
+        StreamingAnomalyDetector instance.
+    """
+    global _streaming_detector
+    if _streaming_detector is None:
+        _streaming_detector = StreamingAnomalyDetector()
+    return _streaming_detector