File size: 7,865 Bytes
59edb07 | 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 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 | """Memory stream — episodic memory with importance scoring and retrieval."""
from __future__ import annotations
import math
from dataclasses import dataclass, field
from enum import Enum
from typing import Optional
class MemoryType(Enum):
OBSERVATION = "observation" # "I saw Maria at the cafe"
REFLECTION = "reflection" # "Maria seems to visit the cafe every morning"
PLAN = "plan" # "I will go to the office at 9am"
CONVERSATION = "conversation" # "I talked to John about the weather"
EVENT = "event" # "A storm hit the city"
@dataclass
class Memory:
"""A single memory entry in an agent's memory stream."""
id: int
tick: int # When this memory was created (simulation tick)
day: int # Day number
time_str: str # Human-readable time "09:15"
type: MemoryType
content: str # Natural language description
importance: int = 5 # 1-10 scale, assigned by LLM
location: str = "" # Where it happened
involved_agents: list[str] = field(default_factory=list) # Other agents involved
# For retrieval scoring
access_count: int = 0
last_accessed_tick: int = 0
def to_dict(self) -> dict:
return {
"id": self.id,
"tick": self.tick,
"day": self.day,
"time_str": self.time_str,
"type": self.type.value,
"content": self.content,
"importance": self.importance,
"location": self.location,
"involved_agents": self.involved_agents,
"access_count": self.access_count,
"last_accessed_tick": self.last_accessed_tick,
}
@classmethod
def from_dict(cls, data: dict) -> Memory:
data = dict(data)
data["type"] = MemoryType(data["type"])
return cls(**data)
class MemoryStream:
"""An agent's full memory — stores, scores, and retrieves memories."""
def __init__(self, max_memories: int = 500) -> None:
self.memories: list[Memory] = []
self.max_memories = max_memories
self._next_id: int = 0
# Running total of importance since last reflection
self._importance_accumulator: float = 0.0
self.reflection_threshold: float = 50.0
def add(
self,
tick: int,
day: int,
time_str: str,
memory_type: MemoryType,
content: str,
importance: int = 5,
location: str = "",
involved_agents: Optional[list[str]] = None,
) -> Memory:
"""Add a new memory to the stream."""
memory = Memory(
id=self._next_id,
tick=tick,
day=day,
time_str=time_str,
type=memory_type,
content=content,
importance=importance,
location=location,
involved_agents=involved_agents or [],
)
self._next_id += 1
self.memories.append(memory)
self._importance_accumulator += importance
# Prune if over capacity — drop lowest-importance, oldest memories
if len(self.memories) > self.max_memories:
self._prune()
return memory
def should_reflect(self) -> bool:
"""True if enough important things have happened to warrant a reflection."""
return self._importance_accumulator >= self.reflection_threshold
def reset_reflection_accumulator(self) -> None:
self._importance_accumulator = 0.0
def retrieve(
self,
current_tick: int,
query: str = "",
top_k: int = 10,
memory_type: Optional[MemoryType] = None,
involved_agent: Optional[str] = None,
) -> list[Memory]:
"""Retrieve top-K most relevant memories using recency + importance scoring.
Score = recency_weight * recency + importance_weight * normalized_importance
For a full implementation, relevance (embedding similarity to query) would be
added as a third factor. For now, we use recency + importance only.
"""
candidates = self.memories
if memory_type:
candidates = [m for m in candidates if m.type == memory_type]
if involved_agent:
candidates = [m for m in candidates if involved_agent in m.involved_agents]
if not candidates:
return []
scored: list[tuple[float, Memory]] = []
for mem in candidates:
recency = self._recency_score(mem.tick, current_tick)
importance = mem.importance / 10.0
# Recency and importance weighted equally
score = 0.5 * recency + 0.5 * importance
scored.append((score, mem))
scored.sort(key=lambda x: x[0], reverse=True)
results = [mem for _, mem in scored[:top_k]]
# Update access tracking
for mem in results:
mem.access_count += 1
mem.last_accessed_tick = current_tick
return results
def get_recent(self, n: int = 5) -> list[Memory]:
"""Get the N most recent memories."""
return self.memories[-n:]
def get_memories_about(self, agent_id: str, top_k: int = 5) -> list[Memory]:
"""Get memories involving a specific agent, most recent first."""
relevant = [m for m in self.memories if agent_id in m.involved_agents]
return relevant[-top_k:]
def get_todays_plan(self, current_day: int) -> list[Memory]:
"""Get today's plan memories."""
return [
m for m in self.memories
if m.type == MemoryType.PLAN and m.day == current_day
]
def _recency_score(self, memory_tick: int, current_tick: int) -> float:
"""Exponential decay based on how many ticks ago the memory was formed."""
age = current_tick - memory_tick
# Decay factor: half-life of ~50 ticks (~12 hours at 15-min ticks)
return math.exp(-0.014 * age)
def _prune(self) -> None:
"""Remove least important, oldest memories when over capacity."""
# Keep reflections and high-importance memories longer
self.memories.sort(
key=lambda m: (
m.type == MemoryType.REFLECTION, # Reflections last
m.importance,
m.tick,
)
)
# Remove the bottom 10%
cut = max(1, len(self.memories) - self.max_memories)
self.memories = self.memories[cut:]
# Re-sort by tick (chronological)
self.memories.sort(key=lambda m: m.tick)
def context_summary(self, current_tick: int, max_memories: int = 15) -> str:
"""Generate a context string of relevant memories for LLM prompts."""
recent = self.retrieve(current_tick, top_k=max_memories)
if not recent:
return "No significant memories yet."
lines = []
for mem in recent:
prefix = f"[Day {mem.day} {mem.time_str}]"
lines.append(f"{prefix} ({mem.type.value}) {mem.content}")
return "\n".join(lines)
def to_dict(self) -> dict:
return {
"memories": [m.to_dict() for m in self.memories],
"next_id": self._next_id,
"importance_accumulator": self._importance_accumulator,
"reflection_threshold": self.reflection_threshold,
"max_memories": self.max_memories,
}
@classmethod
def from_dict(cls, data: dict) -> MemoryStream:
stream = cls(max_memories=data.get("max_memories", 500))
stream._next_id = data["next_id"]
stream._importance_accumulator = data["importance_accumulator"]
stream.reflection_threshold = data.get("reflection_threshold", 50.0)
for md in data["memories"]:
stream.memories.append(Memory.from_dict(md))
return stream
|