File size: 2,767 Bytes
d2a5f5a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 | """
Attentional Gate — Vitalis FSI
Controls what gets through to the cognitive core.
High-salience inputs pass. Low-salience inputs are filtered.
Salience = novelty * valence_magnitude * relevance_to_identity
Biological analog: thalamic gating — the thalamus doesn't
pass everything to cortex, only what matters right now.
"""
import numpy as np
from vitalis_ide.math_core.kernel import VitalisKernel
from src.valence.valence_engine import ValenceEngine
class AttentionalGate:
SALIENCE_THRESHOLD = 0.15
IDENTITY_WEIGHT = 0.3
NOVELTY_WEIGHT = 0.4
VALENCE_WEIGHT = 0.3
def __init__(self, valence: ValenceEngine, identity_vec: np.ndarray = None):
self.kernel = VitalisKernel()
self.valence = valence
self.identity_vec = identity_vec
self._recent = []
self._recent_max = 20
self._passed = 0
self._filtered = 0
def set_identity(self, vec: np.ndarray):
self.identity_vec = vec
def _novelty(self, hv: np.ndarray) -> float:
if not self._recent:
return 1.0
sims = [self.kernel.similarity(hv, r) for r in self._recent]
return float(1.0 - max(sims))
def _identity_relevance(self, hv: np.ndarray) -> float:
if self.identity_vec is None:
return 0.5
return float(max(0.0, self.kernel.similarity(hv, self.identity_vec)))
def gate(self, hv: np.ndarray, force: bool = False) -> tuple:
"""
Returns (passes: bool, salience: float, reason: str)
"""
novelty = self._novelty(hv)
val, vconf = self.valence.evaluate(hv)
relevance = self._identity_relevance(hv)
salience = (
self.NOVELTY_WEIGHT * novelty +
self.VALENCE_WEIGHT * vconf +
self.IDENTITY_WEIGHT * relevance
)
salience = float(np.clip(salience, 0.0, 1.0))
passes = force or salience >= self.SALIENCE_THRESHOLD
if passes:
self._recent.append(hv.copy())
if len(self._recent) > self._recent_max:
self._recent.pop(0)
self._passed += 1
reason = f"salience={salience:.3f} novelty={novelty:.3f} relevance={relevance:.3f}"
else:
self._filtered += 1
reason = f"filtered salience={salience:.3f} below threshold={self.SALIENCE_THRESHOLD}"
return passes, salience, reason
def report(self) -> dict:
total = self._passed + self._filtered
return {
"passed": self._passed,
"filtered": self._filtered,
"pass_rate": round(self._passed / total, 3) if total else 0.0,
"recent_inputs": len(self._recent),
}
|