superlocalmemory 3.2.2 → 3.3.0

package/src/superlocalmemory/dynamics/ebbinghaus_langevin_coupling.py

@@ -0,0 +1,171 @@
+# Copyright (c) 2026 Varun Pratap Bhardwaj / Qualixar
+# Licensed under the MIT License - see LICENSE file
+# Part of SuperLocalMemory V3 | https://qualixar.com | https://varunpratap.com
+
+"""Ebbinghaus-Langevin coupling — forgetting drift in dynamics.
+
+Combines the Ebbinghaus forgetting curve with Fisher-Langevin coupling
+to create a three-way information-dynamic lifecycle:
+
+  1. Fisher confidence  -> Langevin temperature (existing)
+  2. Ebbinghaus retention -> forgetting drift coefficient (new)
+  3. Combined temperature = T_fisher * (1 + lambda_forget)
+
+The forgetting drift pushes low-retention memories toward the Langevin
+boundary faster, while high-retention memories resist drift. This creates
+a thermodynamically grounded forgetting process.
+
+Mathematical formulation:
+  lambda_forget = (1 - R) * forgetting_drift_scale
+  T_combined = T_fisher * (1 + lambda_forget)
+  weight_combined = w_fisher * w_ebbinghaus
+
+Part of Qualixar | Author: Varun Pratap Bhardwaj
+License: MIT
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+
+import numpy as np
+
+from superlocalmemory.core.config import ForgettingConfig
+from superlocalmemory.dynamics.fisher_langevin_coupling import (
+    FisherLangevinCoupling,
+)
+from superlocalmemory.math.ebbinghaus import EbbinghausCurve
+from superlocalmemory.math.langevin import LangevinDynamics
+
+logger = logging.getLogger(__name__)
+
+
+# ---------------------------------------------------------------------------
+# Coupling state
+# ---------------------------------------------------------------------------
+
+@dataclass(frozen=True)
+class EbbinghausCouplingState:
+    """State of the Ebbinghaus-Langevin coupling for a single memory.
+
+    Attributes:
+        fact_id: Fact identifier.
+        retention_score: R(t) in [0, 1].
+        memory_strength: S(m) in [S_MIN, S_MAX].
+        lifecycle_zone: One of active/warm/cold/archive/forgotten.
+        effective_temperature: Combined T (Fisher + forgetting).
+        lifecycle_weight: Combined weight [0, 1].
+        forgetting_drift: lambda_forget coefficient.
+        is_forgotten: Below forget_threshold.
+        consolidation_signal: From Fisher coupling.
+    """
+
+    fact_id: str
+    retention_score: float
+    memory_strength: float
+    lifecycle_zone: str
+    effective_temperature: float
+    lifecycle_weight: float
+    forgetting_drift: float
+    is_forgotten: bool
+    consolidation_signal: bool
+
+
+# ---------------------------------------------------------------------------
+# Coupling class
+# ---------------------------------------------------------------------------
+
+class EbbinghausLangevinCoupling:
+    """Couples Ebbinghaus forgetting to Fisher-Langevin dynamics.
+
+    Creates a three-way feedback loop:
+    1. Fisher confidence modulates Langevin temperature (existing).
+    2. Ebbinghaus retention adds forgetting drift.
+    3. Combined effect determines lifecycle zone and weight.
+    """
+
+    __slots__ = ("_ebbinghaus", "_langevin", "_fisher_coupling", "_config")
+
+    def __init__(
+        self,
+        ebbinghaus: EbbinghausCurve,
+        langevin: LangevinDynamics,
+        fisher_coupling: FisherLangevinCoupling,
+        config: ForgettingConfig,
+    ) -> None:
+        self._ebbinghaus = ebbinghaus
+        self._langevin = langevin
+        self._fisher_coupling = fisher_coupling
+        self._config = config
+
+    def compute_coupled_state(
+        self,
+        fact_id: str,
+        fisher_variance: np.ndarray,
+        langevin_radius: float,
+        access_count: int,
+        importance: float,
+        confirmation_count: int,
+        emotional_salience: float,
+        hours_since_last_access: float,
+    ) -> EbbinghausCouplingState:
+        """Compute the full coupled state for a single memory.
+
+        Combines Fisher-Langevin coupling with Ebbinghaus forgetting
+        to produce a unified lifecycle state.
+
+        Args:
+            fact_id: Fact identifier.
+            fisher_variance: Fisher variance vector (diagonal).
+            langevin_radius: Current Langevin position radius [0, 1).
+            access_count: Total access count.
+            importance: PageRank importance score.
+            confirmation_count: Evidence/confirmation count.
+            emotional_salience: Emotional strength.
+            hours_since_last_access: Hours since last access.
+
+        Returns:
+            EbbinghausCouplingState with all computed fields.
+        """
+        # Step 1: Fisher-Langevin coupling (existing system)
+        fl_state = self._fisher_coupling.compute_coupling(
+            fisher_variance, langevin_radius, access_count,
+        )
+
+        # Step 2: Ebbinghaus strength
+        strength = self._ebbinghaus.memory_strength(
+            access_count, importance, confirmation_count, emotional_salience,
+        )
+
+        # Step 3: Ebbinghaus retention
+        retention = self._ebbinghaus.retention(hours_since_last_access, strength)
+
+        # Step 4: Lifecycle zone
+        zone = self._ebbinghaus.lifecycle_zone(retention)
+
+        # Step 5: Forgetting drift coefficient
+        # Higher forgetting (lower R) -> stronger drift toward boundary
+        lambda_forget = (1.0 - retention) * self._config.forgetting_drift_scale
+
+        # Step 6: Combined temperature
+        # Forgetting increases effective temperature -> more noise -> faster drift
+        t_combined = fl_state.langevin_temperature * (1.0 + lambda_forget)
+
+        # Step 7: Combined weight
+        weight = fl_state.lifecycle_weight * self._ebbinghaus.lifecycle_weight(zone)
+
+        # Step 8: Is forgotten?
+        is_forgotten = zone == "forgotten"
+
+        return EbbinghausCouplingState(
+            fact_id=fact_id,
+            retention_score=retention,
+            memory_strength=strength,
+            lifecycle_zone=zone,
+            effective_temperature=t_combined,
+            lifecycle_weight=weight,
+            forgetting_drift=lambda_forget,
+            is_forgotten=is_forgotten,
+            consolidation_signal=fl_state.consolidation_signal,
+        )