liquid-loop 0.6.0__tar.gz → 0.6.1__tar.gz

{liquid_loop-0.6.0 → liquid_loop-0.6.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: liquid-loop
-Version: 0.6.0
+Version: 0.6.1
 Summary: Self-Organizing Cognitive Memory for AI Agents — Liquid Loop theory implementation
 Author: fishbook0001
 Maintainer: fishbook0001

{liquid_loop-0.6.0 → liquid_loop-0.6.1}/liquid_loop/__init__.py

@@ -1,9 +1,10 @@
-"""Liquid Loop — Workspace Cognitive Runtime v0.6.0 (CPE-fused, proactive)"""
-__version__ = "0.6.0"
+"""Liquid Loop — Workspace Cognitive Runtime v0.6.1 (CPE-fused, proactive, rhythmic sampling)"""
+__version__ = "0.6.1"
 
 from .workspace import (
     WorkspaceState, Anchor, Evidence, Memory, Conflict,
     AuditChain, CPERegularizer, SelfRefineEngine,
+    rhythmic_retrieve,
 )
 from .storage import load, save
 from .entropy import calculate, calculate_detail, calculate as calculate_entropy

{liquid_loop-0.6.0 → liquid_loop-0.6.1}/liquid_loop/storage.py

@@ -57,6 +57,8 @@ def save(state: WorkspaceState, workspace_root: Path):
     state.audit_prev_hash = state.audit_chain_hash
     state.audit_chain_hash = audit_hash
     data = asdict(state)
+    # overlap_cache 仅为运行时熵计算缓存，键为 tuple，不可 JSON 序列化，且不具持久价值
+    data.pop("overlap_cache", None)
     with open(path, "w", encoding="utf-8") as f:
         json.dump(data, f, indent=2, ensure_ascii=False)
 

{liquid_loop-0.6.0 → liquid_loop-0.6.1}/liquid_loop/workspace.py

@@ -658,7 +658,8 @@ def _keyword_overlap(a: str, b: str, cache: dict | None = None) -> float:
         return 0.0
     if cache is not None:
         import hashlib
-        key = tuple(sorted([hashlib.md5(a.encode()).hexdigest()[:8], hashlib.md5(b.encode()).hexdigest()[:8]]))
+        h1, h2 = hashlib.md5(a.encode()).hexdigest()[:8], hashlib.md5(b.encode()).hexdigest()[:8]
+        key = f"{min(h1,h2)}:{max(h1,h2)}"  # str key，JSON 兼容
         if key in cache:
             return cache[key]
     import re
@@ -896,3 +897,61 @@ def meta_thinker_advice(anchor: Anchor, state: WorkspaceState, new_evidence: str
         return {"action": "ADD", "reason": f"弱相关(overlap={best_score:.2f})，添加但标记为low value"}
     else:
         return {"action": "SKIP", "reason": "与当前锚点无关联"}
+
+
+# ==============================================================================
+# 节律采样检索（Rhythmic Sampling Retrieve）
+# 启发来源：Biba et al. 2026, Nature Human Behaviour — 7Hz theta 脉冲记忆编码
+# 思路：记忆检索不是"取 top-N"，而是分窗口脉冲采样，每组取最优，跨组去重
+# ==============================================================================
+
+def rhythmic_retrieve(
+    state: WorkspaceState,
+    query: str,
+    window_size: int = 7,
+    top_per_window: int = 1,
+    total_slots: int = 5,
+) -> list[str]:
+    """节律采样检索：分窗口脉冲采样，每组取最优，跨组去重。
+
+    - window_size: 每个"脉冲窗口"的候选数量（默认 7，呼应 7Hz theta 节律）
+    - top_per_window: 每个窗口保留的条数
+    - total_slots: 最终返回的总条数
+
+    比直接 top-N 的优势：避免同质记忆堆叠，增加多样性
+    """
+    if not query:
+        return []
+
+    # 候选集：证据 + 锚点描述
+    candidates: list[str] = []
+    for e in state.evidences:
+        if e.content:
+            candidates.append(e.content)
+    for a in state.anchors:
+        if a.description:
+            candidates.append(a.description)
+        if a.name:
+            candidates.append(a.name)
+
+    if not candidates:
+        return []
+
+    # 计算 overlap 并排序
+    scored = [(_keyword_overlap(query, c, state.overlap_cache), c) for c in candidates]
+    scored.sort(key=lambda x: x[0], reverse=True)
+
+    # 节律窗口采样：每 window_size 个为一组脉冲，每组取 top_per_window
+    result = []
+    seen = set()
+    for i in range(0, len(scored), window_size):
+        window = scored[i : i + window_size]
+        # 每组内按 overlap 降序取 top_per_window
+        for _, content in window[:top_per_window]:
+            if content not in seen and len(result) < total_slots:
+                seen.add(content)
+                result.append(content)
+        if len(result) >= total_slots:
+            break
+
+    return result[:total_slots]

{liquid_loop-0.6.0 → liquid_loop-0.6.1}/liquid_loop.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: liquid-loop
-Version: 0.6.0
+Version: 0.6.1
 Summary: Self-Organizing Cognitive Memory for AI Agents — Liquid Loop theory implementation
 Author: fishbook0001
 Maintainer: fishbook0001

{liquid_loop-0.6.0 → liquid_loop-0.6.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "liquid-loop"
-version = "0.6.0"
+version = "0.6.1"
 description = "Self-Organizing Cognitive Memory for AI Agents — Liquid Loop theory implementation"
 readme = "README.md"
 license = {text = "MIT"}