oscura 0.0.1__py3-none-any.whl → 0.1.1__py3-none-any.whl

oscura/automotive/can/patterns.py

@@ -0,0 +1,381 @@
+"""Multi-message pattern learning for CAN bus analysis.
+
+This module provides algorithms for discovering relationships between CAN messages,
+including message pairs, sequences, and temporal correlations.
+"""
+
+from __future__ import annotations
+
+from collections import defaultdict
+from dataclasses import dataclass
+from typing import TYPE_CHECKING
+
+import numpy as np
+
+if TYPE_CHECKING:
+    from oscura.automotive.can.session import CANSession
+
+__all__ = [
+    "MessagePair",
+    "MessageSequence",
+    "PatternAnalyzer",
+    "TemporalCorrelation",
+]
+
+
+@dataclass
+class MessagePair:
+    """A pair of CAN messages that frequently occur together.
+
+    Attributes:
+        id_a: First message arbitration ID.
+        id_b: Second message arbitration ID.
+        occurrences: Number of times this pair occurred within time window.
+        avg_delay_ms: Average time delay from id_a to id_b in milliseconds.
+        confidence: Confidence score (0.0-1.0) based on consistency.
+    """
+
+    id_a: int
+    id_b: int
+    occurrences: int
+    avg_delay_ms: float
+    confidence: float
+
+    def __repr__(self) -> str:
+        """Human-readable representation."""
+        return (
+            f"MessagePair(0x{self.id_a:03X} → 0x{self.id_b:03X}, "
+            f"occurrences={self.occurrences}, "
+            f"delay={self.avg_delay_ms:.2f}ms, "
+            f"confidence={self.confidence:.2f})"
+        )
+
+
+@dataclass
+class MessageSequence:
+    """A sequence of CAN messages that occur in order.
+
+    Attributes:
+        ids: List of message arbitration IDs in sequence order.
+        occurrences: Number of times this sequence occurred.
+        avg_timing: Average delays between consecutive messages in milliseconds.
+        support: Support score (0.0-1.0) - fraction of time sequence appears.
+    """
+
+    ids: list[int]
+    occurrences: int
+    avg_timing: list[float]
+    support: float
+
+    def __repr__(self) -> str:
+        """Human-readable representation."""
+        id_str = " → ".join(f"0x{id_:03X}" for id_ in self.ids)
+        timing_str = " → ".join(f"{t:.1f}ms" for t in self.avg_timing)
+        return (
+            f"MessageSequence({id_str}, "
+            f"occurrences={self.occurrences}, "
+            f"timing=[{timing_str}], "
+            f"support={self.support:.2f})"
+        )
+
+
+@dataclass
+class TemporalCorrelation:
+    """Temporal correlation between two CAN messages.
+
+    Attributes:
+        leader_id: Message ID that typically appears first.
+        follower_id: Message ID that typically appears after leader.
+        avg_delay_ms: Average delay from leader to follower in milliseconds.
+        std_delay_ms: Standard deviation of delay in milliseconds.
+        occurrences: Number of times this correlation was observed.
+    """
+
+    leader_id: int
+    follower_id: int
+    avg_delay_ms: float
+    std_delay_ms: float
+    occurrences: int
+
+    def __repr__(self) -> str:
+        """Human-readable representation."""
+        return (
+            f"TemporalCorrelation(0x{self.leader_id:03X} → 0x{self.follower_id:03X}, "
+            f"delay={self.avg_delay_ms:.2f}±{self.std_delay_ms:.2f}ms, "
+            f"occurrences={self.occurrences})"
+        )
+
+
+class PatternAnalyzer:
+    """Detect patterns in CAN message sequences.
+
+    This class provides static methods for discovering relationships between
+    CAN messages, useful for understanding message dependencies and control flows.
+
+    Example - Find message pairs:
+        >>> session = CANSession.from_log("capture.blf")
+        >>> pairs = PatternAnalyzer.find_message_pairs(session, time_window_ms=100)
+        >>> for pair in pairs:
+        ...     print(pair)
+
+    Example - Find message sequences:
+        >>> sequences = PatternAnalyzer.find_message_sequences(
+        ...     session,
+        ...     max_sequence_length=3,
+        ...     time_window_ms=500
+        ... )
+
+    Example - Find temporal correlations:
+        >>> correlations = PatternAnalyzer.find_temporal_correlations(
+        ...     session,
+        ...     max_delay_ms=100
+        ... )
+    """
+
+    @staticmethod
+    def find_message_pairs(
+        session: CANSession,
+        time_window_ms: float = 100,
+        min_occurrence: int = 3,
+    ) -> list[MessagePair]:
+        """Find message pairs that frequently occur together.
+
+        Uses a sliding time window to detect messages that consistently appear
+        within a short time of each other. This can reveal request-response
+        patterns or coordinated control messages.
+
+        Args:
+            session: CAN session to analyze.
+            time_window_ms: Maximum time window in milliseconds to consider
+                messages as a pair.
+            min_occurrence: Minimum number of occurrences to report a pair.
+
+        Returns:
+            List of MessagePair objects, sorted by occurrence count (descending).
+        """
+        time_window_s = time_window_ms / 1000.0
+
+        # Get all messages sorted by timestamp
+        all_messages = sorted(session._messages.messages, key=lambda m: m.timestamp)
+
+        # Track pairs: (id_a, id_b) -> list of delays
+        pair_delays: dict[tuple[int, int], list[float]] = defaultdict(list)
+
+        # Use sliding window approach
+        for i, msg_a in enumerate(all_messages):
+            # Look forward in time window
+            for msg_b in all_messages[i + 1 :]:
+                delay = msg_b.timestamp - msg_a.timestamp
+                if delay > time_window_s:
+                    break  # Outside window, stop searching
+
+                # Don't pair message with itself (same ID)
+                if msg_a.arbitration_id != msg_b.arbitration_id:
+                    pair_key = (msg_a.arbitration_id, msg_b.arbitration_id)
+                    pair_delays[pair_key].append(delay * 1000)  # Convert to ms
+
+        # Build MessagePair objects
+        pairs = []
+        for (id_a, id_b), delays in pair_delays.items():
+            occurrences = len(delays)
+            if occurrences >= min_occurrence:
+                avg_delay = np.mean(delays)
+                std_delay = np.std(delays)
+
+                # Confidence based on consistency (lower std = higher confidence)
+                # Normalize std by average delay
+                if avg_delay > 0:
+                    cv = std_delay / avg_delay  # Coefficient of variation
+                    confidence = float(1.0 / (1.0 + cv))  # Higher confidence for lower CV
+                else:
+                    confidence = 1.0
+
+                pairs.append(
+                    MessagePair(
+                        id_a=id_a,
+                        id_b=id_b,
+                        occurrences=occurrences,
+                        avg_delay_ms=float(avg_delay),
+                        confidence=float(confidence),
+                    )
+                )
+
+        # Sort by occurrence count (descending)
+        pairs.sort(key=lambda p: p.occurrences, reverse=True)
+
+        return pairs
+
+    @staticmethod
+    def find_message_sequences(
+        session: CANSession,
+        max_sequence_length: int = 5,
+        time_window_ms: float = 500,
+        min_support: float = 0.7,
+    ) -> list[MessageSequence]:
+        """Find message sequences (A → B → C patterns).
+
+        Uses sequential pattern mining to discover message sequences that
+        frequently occur in order. This can reveal multi-step control
+        sequences or protocol handshakes.
+
+        Args:
+            session: CAN session to analyze.
+            max_sequence_length: Maximum length of sequences to find (2-10).
+            time_window_ms: Maximum time window for entire sequence.
+            min_support: Minimum support (0.0-1.0) - fraction of occurrences
+                relative to most frequent message.
+
+        Returns:
+            List of MessageSequence objects, sorted by support (descending).
+
+        Raises:
+            ValueError: If max_sequence_length is invalid (<2 or >10) or
+                min_support is not in range [0.0, 1.0].
+        """
+        if max_sequence_length < 2:
+            raise ValueError("max_sequence_length must be at least 2")
+        if max_sequence_length > 10:
+            raise ValueError("max_sequence_length cannot exceed 10")
+        if not 0.0 <= min_support <= 1.0:
+            raise ValueError("min_support must be between 0.0 and 1.0")
+
+        time_window_s = time_window_ms / 1000.0
+
+        # Get all messages sorted by timestamp
+        all_messages = sorted(session._messages.messages, key=lambda m: m.timestamp)
+
+        if not all_messages:
+            return []
+
+        # Find maximum message frequency (for support calculation)
+        id_counts: dict[int, int] = defaultdict(int)
+        for msg in all_messages:
+            id_counts[msg.arbitration_id] += 1
+        max_count = max(id_counts.values()) if id_counts else 1
+
+        # Mine sequences of increasing length
+        sequences: dict[tuple[int, ...], list[list[float]]] = defaultdict(list)
+
+        # Start with 2-message sequences
+        for i, msg_a in enumerate(all_messages):
+            # Look forward in time window
+            sequence_start = msg_a.timestamp
+            current_sequence = [msg_a.arbitration_id]
+            timing = []
+
+            for msg_b in all_messages[i + 1 :]:
+                if msg_b.timestamp - sequence_start > time_window_s:
+                    break
+
+                # Extend sequence
+                current_sequence.append(msg_b.arbitration_id)
+                timing.append((msg_b.timestamp - msg_a.timestamp) * 1000)  # ms
+
+                # Record sequences of desired length
+                if 2 <= len(current_sequence) <= max_sequence_length:
+                    seq_key = tuple(current_sequence)
+                    sequences[seq_key].append(timing.copy())
+
+                # Update for next iteration
+                msg_a = msg_b
+                if len(current_sequence) >= max_sequence_length:
+                    break
+
+        # Build MessageSequence objects
+        result = []
+        for seq_ids, timing_lists in sequences.items():
+            occurrences = len(timing_lists)
+            support = occurrences / max_count
+
+            if support >= min_support:
+                # Calculate average timing between consecutive messages
+                avg_timing = []
+                if timing_lists:
+                    # timing_lists[i] has len(seq_ids) - 1 elements
+                    num_gaps = len(seq_ids) - 1
+                    for gap_idx in range(num_gaps):
+                        gap_times = [t[gap_idx] for t in timing_lists if gap_idx < len(t)]
+                        if gap_times:
+                            avg_timing.append(float(np.mean(gap_times)))
+
+                result.append(
+                    MessageSequence(
+                        ids=list(seq_ids),
+                        occurrences=occurrences,
+                        avg_timing=avg_timing,
+                        support=float(support),
+                    )
+                )
+
+        # Sort by support (descending), then by occurrences
+        result.sort(key=lambda s: (s.support, s.occurrences), reverse=True)
+
+        return result
+
+    @staticmethod
+    def find_temporal_correlations(
+        session: CANSession,
+        max_delay_ms: float = 100,
+    ) -> dict[tuple[int, int], TemporalCorrelation]:
+        """Find temporal correlations between messages.
+
+        Analyzes timing relationships to determine which messages consistently
+        follow others with predictable delays. Unlike message pairs, this focuses
+        on statistical correlation rather than simple co-occurrence.
+
+        Args:
+            session: CAN session to analyze.
+            max_delay_ms: Maximum delay to consider for correlations.
+
+        Returns:
+            Dictionary mapping (leader_id, follower_id) to TemporalCorrelation.
+        """
+        max_delay_s = max_delay_ms / 1000.0
+
+        # Get all messages sorted by timestamp
+        all_messages = sorted(session._messages.messages, key=lambda m: m.timestamp)
+
+        # Track correlations: (leader_id, follower_id) -> list of delays
+        correlation_delays: dict[tuple[int, int], list[float]] = defaultdict(list)
+
+        # For each message, find the next occurrence of each other message ID
+        # within the time window
+        for i, leader_msg in enumerate(all_messages):
+            leader_id = leader_msg.arbitration_id
+
+            # Track which follower IDs we've seen (only count first occurrence)
+            seen_followers = set()
+
+            # Look forward for followers
+            for follower_msg in all_messages[i + 1 :]:
+                delay = follower_msg.timestamp - leader_msg.timestamp
+                if delay > max_delay_s:
+                    break  # Outside window
+
+                follower_id = follower_msg.arbitration_id
+
+                # Skip same ID
+                if follower_id == leader_id:
+                    continue
+
+                # Only record first occurrence of each follower ID
+                if follower_id not in seen_followers:
+                    correlation_delays[(leader_id, follower_id)].append(delay * 1000)
+                    seen_followers.add(follower_id)
+
+        # Build TemporalCorrelation objects
+        correlations = {}
+        for (leader_id, follower_id), delays in correlation_delays.items():
+            if len(delays) >= 2:  # Need at least 2 samples for std
+                avg_delay = np.mean(delays)
+                std_delay = np.std(delays)
+
+                correlations[(leader_id, follower_id)] = TemporalCorrelation(
+                    leader_id=leader_id,
+                    follower_id=follower_id,
+                    avg_delay_ms=float(avg_delay),
+                    std_delay_ms=float(std_delay),
+                    occurrences=len(delays),
+                )
+
+        return correlations