MaiBot/src/chat/frequency_control/frequency_control.py

508 lines
22 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

import time
from typing import Optional, Dict, List
from src.plugin_system.apis import message_api
from src.chat.message_receive.chat_stream import ChatStream, get_chat_manager
from src.common.logger import get_logger
from src.config.config import global_config
from src.chat.frequency_control.talk_frequency_control import get_config_base_talk_frequency
from src.chat.frequency_control.focus_value_control import get_config_base_focus_value
logger = get_logger("frequency_control")
class FrequencyControl:
"""
频率控制类,可以根据最近时间段的发言数量和发言人数动态调整频率
特点:
- 发言频率调整基于最近10分钟的数据评估单位为"消息数/10分钟"
- 专注度调整基于最近10分钟的数据评估单位为"消息数/10分钟"
- 历史基准值基于最近一周的数据按小时统计每小时都有独立的基准值需要至少50条历史消息
- 统一标准两个调整都使用10分钟窗口确保逻辑一致性和响应速度
- 双向调整:根据活跃度高低,既能提高也能降低频率和专注度
- 数据充足性检查当历史数据不足50条时不更新基准值当基准值为默认值时不进行动态调整
- 基准值更新:直接使用新计算的周均值,无平滑更新
"""
def __init__(self, chat_id: str):
self.chat_id = chat_id
self.chat_stream: ChatStream = get_chat_manager().get_stream(self.chat_id) # type: ignore
if not self.chat_stream:
raise ValueError(f"无法找到聊天流: {chat_id}")
self.log_prefix = f"[{get_chat_manager().get_stream_name(self.chat_id) or self.chat_id}]"
# 发言频率调整值
self.talk_frequency_adjust: float = 1.0
self.talk_frequency_external_adjust: float = 1.0
# 专注度调整值
self.focus_value_adjust: float = 1.0
self.focus_value_external_adjust: float = 1.0
# 动态调整相关参数
self.last_update_time = time.time()
self.update_interval = 60 # 每60秒更新一次
# 历史数据缓存
self._message_count_cache = 0
self._user_count_cache = 0
self._last_cache_time = 0
self._cache_duration = 30 # 缓存30秒
# 调整参数
self.min_adjust = 0.3 # 最小调整值
self.max_adjust = 2.0 # 最大调整值
# 动态基准值(将根据历史数据计算)
self.base_message_count = 5 # 默认基准消息数量,将被动态更新
self.base_user_count = 3 # 默认基准用户数量,将被动态更新
# 平滑因子
self.smoothing_factor = 0.3
# 历史数据相关参数
self._last_historical_update = 0
self._historical_update_interval = 600 # 每十分钟更新一次历史基准值
self._historical_days = 7 # 使用最近7天的数据计算基准值
# 按小时统计的历史基准值
self._hourly_baseline = {
"messages": {}, # {0-23: 平均消息数}
"users": {}, # {0-23: 平均用户数}
}
# 初始化24小时的默认基准值
for hour in range(24):
self._hourly_baseline["messages"][hour] = 0.0
self._hourly_baseline["users"][hour] = 0.0
def _update_historical_baseline(self):
"""
更新基于历史数据的基准值
使用最近一周的数据,按小时统计平均消息数量和用户数量
"""
current_time = time.time()
# 检查是否需要更新历史基准值
if current_time - self._last_historical_update < self._historical_update_interval:
return
try:
# 计算一周前的时间戳
week_ago = current_time - (self._historical_days * 24 * 3600)
# 获取最近一周的消息数据
historical_messages = message_api.get_messages_by_time_in_chat(
chat_id=self.chat_stream.stream_id,
start_time=week_ago,
end_time=current_time,
filter_mai=True,
filter_command=True,
)
if historical_messages and len(historical_messages) >= 50:
# 按小时统计消息数和用户数
hourly_stats = {hour: {"messages": [], "users": set()} for hour in range(24)}
for msg in historical_messages:
# 获取消息的小时UTC时间
msg_time = time.localtime(msg.time)
msg_hour = msg_time.tm_hour
# 统计消息数
hourly_stats[msg_hour]["messages"].append(msg)
# 统计用户数
if msg.user_info and msg.user_info.user_id:
hourly_stats[msg_hour]["users"].add(msg.user_info.user_id)
# 计算每个小时的平均值(基于一周的数据)
for hour in range(24):
# 计算该小时的平均消息数(一周内该小时的总消息数 / 7天
total_messages = len(hourly_stats[hour]["messages"])
total_users = len(hourly_stats[hour]["users"])
# 只计算有消息的时段没有消息的时段设为0
if total_messages > 0:
avg_messages = total_messages / self._historical_days
avg_users = total_users / self._historical_days
self._hourly_baseline["messages"][hour] = avg_messages
self._hourly_baseline["users"][hour] = avg_users
else:
# 没有消息的时段设为0表示该时段不活跃
self._hourly_baseline["messages"][hour] = 0.0
self._hourly_baseline["users"][hour] = 0.0
# 更新整体基准值(用于兼容性)- 基于原始数据计算不受max(1.0)限制影响
overall_avg_messages = sum(len(hourly_stats[hour]["messages"]) for hour in range(24)) / (
24 * self._historical_days
)
overall_avg_users = sum(len(hourly_stats[hour]["users"]) for hour in range(24)) / (
24 * self._historical_days
)
self.base_message_count = overall_avg_messages
self.base_user_count = overall_avg_users
logger.info(
f"{self.log_prefix} 历史基准值更新完成: "
f"整体平均消息数={overall_avg_messages:.2f}, 整体平均用户数={overall_avg_users:.2f}"
)
# 记录几个关键时段的基准值
key_hours = [8, 12, 18, 22] # 早、中、晚、夜
for hour in key_hours:
# 计算该小时平均每10分钟的消息数和用户数
hourly_10min_messages = self._hourly_baseline["messages"][hour] / 6 # 1小时 = 6个10分钟
hourly_10min_users = self._hourly_baseline["users"][hour] / 6
logger.info(
f"{self.log_prefix} {hour}时基准值: "
f"消息数={self._hourly_baseline['messages'][hour]:.2f}/小时 "
f"({hourly_10min_messages:.2f}/10分钟), "
f"用户数={self._hourly_baseline['users'][hour]:.2f}/小时 "
f"({hourly_10min_users:.2f}/10分钟)"
)
elif historical_messages and len(historical_messages) < 50:
# 历史数据不足50条不更新基准值
logger.info(f"{self.log_prefix} 历史数据不足50条({len(historical_messages)}条),不更新基准值")
else:
# 如果没有历史数据,不更新基准值
logger.info(f"{self.log_prefix} 无历史数据,不更新基准值")
except Exception as e:
logger.error(f"{self.log_prefix} 更新历史基准值时出错: {e}")
# 出错时保持原有基准值不变
self._last_historical_update = current_time
def _get_current_hour_baseline(self) -> tuple[float, float]:
"""
获取当前小时的基准值
Returns:
tuple: (基准消息数, 基准用户数)
"""
current_hour = time.localtime().tm_hour
return (self._hourly_baseline["messages"][current_hour], self._hourly_baseline["users"][current_hour])
def get_dynamic_talk_frequency_adjust(self) -> float:
"""
获取纯动态调整值(不包含配置文件基础值)
Returns:
float: 动态调整值
"""
self._update_talk_frequency_adjust()
return self.talk_frequency_adjust
def get_dynamic_focus_value_adjust(self) -> float:
"""
获取纯动态调整值(不包含配置文件基础值)
Returns:
float: 动态调整值
"""
self._update_focus_value_adjust()
return self.focus_value_adjust
def _update_talk_frequency_adjust(self):
"""
更新发言频率调整值
适合人少话多的时候:人少但消息多,提高回复频率
"""
current_time = time.time()
# 检查是否需要更新
if current_time - self.last_update_time < self.update_interval:
return
# 先更新历史基准值
self._update_historical_baseline()
try:
# 获取最近10分钟的数据发言频率更敏感
recent_messages = message_api.get_messages_by_time_in_chat(
chat_id=self.chat_stream.stream_id,
start_time=current_time - 600, # 10分钟前
end_time=current_time,
filter_mai=True,
filter_command=True,
)
# 计算消息数量和用户数量
message_count = len(recent_messages)
user_ids = set()
for msg in recent_messages:
if msg.user_info and msg.user_info.user_id:
user_ids.add(msg.user_info.user_id)
user_count = len(user_ids)
# 获取当前小时的基准值
current_hour_base_messages, current_hour_base_users = self._get_current_hour_baseline()
# 计算当前小时平均每10分钟的基准值
current_hour_10min_messages = current_hour_base_messages / 6 # 1小时 = 6个10分钟
current_hour_10min_users = current_hour_base_users / 6
# 发言频率调整逻辑:根据活跃度双向调整
# 检查是否有足够的数据进行分析
if user_count > 0 and message_count >= 2: # 至少需要2条消息才能进行有意义的分析
# 检查历史基准值是否有效(该时段有活跃度)
if current_hour_base_messages > 0.0 and current_hour_base_users > 0.0:
# 计算人均消息数10分钟窗口
messages_per_user = message_count / user_count
# 使用当前小时每10分钟的基准人均消息数
base_messages_per_user = (
current_hour_10min_messages / current_hour_10min_users if current_hour_10min_users > 0 else 1.0
)
# 双向调整逻辑
if messages_per_user > base_messages_per_user * 1.2:
# 活跃度很高:提高回复频率
target_talk_adjust = min(self.max_adjust, messages_per_user / base_messages_per_user)
elif messages_per_user < base_messages_per_user * 0.8:
# 活跃度很低:降低回复频率
target_talk_adjust = max(self.min_adjust, messages_per_user / base_messages_per_user)
else:
# 活跃度正常:保持正常
target_talk_adjust = 1.0
else:
# 历史基准值不足,不调整
target_talk_adjust = 1.0
else:
# 数据不足:不调整
target_talk_adjust = 1.0
# 限制调整范围
target_talk_adjust = max(self.min_adjust, min(self.max_adjust, target_talk_adjust))
# 记录调整前的值
old_adjust = self.talk_frequency_adjust
# 平滑调整
self.talk_frequency_adjust = (
self.talk_frequency_adjust * (1 - self.smoothing_factor) + target_talk_adjust * self.smoothing_factor
)
# 判断调整方向
if target_talk_adjust > 1.0:
adjust_direction = "提高"
elif target_talk_adjust < 1.0:
adjust_direction = "降低"
else:
if current_hour_base_messages <= 0.0 or current_hour_base_users <= 0.0:
adjust_direction = "不调整(该时段无活跃度)"
else:
adjust_direction = "保持"
# 计算实际变化方向
actual_change = ""
if self.talk_frequency_adjust > old_adjust:
actual_change = f"{old_adjust:.2f}x → {self.talk_frequency_adjust:.2f}x"
elif self.talk_frequency_adjust < old_adjust:
actual_change = f"{old_adjust:.2f}x → {self.talk_frequency_adjust:.2f}x"
else:
actual_change = f"无变化: {self.talk_frequency_adjust:.2f}x"
logger.info(
f"{self.log_prefix} 发言频率调整: "
f"{user_count}名用户正在参与聊天,当前消息数: {message_count}|"
f"群基准: {current_hour_10min_messages:.2f}消息/{current_hour_10min_users:.2f}用户|"
f"[{adjust_direction}]{actual_change}"
)
except Exception as e:
logger.error(f"{self.log_prefix} 更新发言频率调整值时出错: {e}")
def _update_focus_value_adjust(self):
"""
更新专注度调整值
适合人多话多的时候人多且消息多提高专注度LLM消耗更多但回复更精准
"""
current_time = time.time()
# 检查是否需要更新
if current_time - self.last_update_time < self.update_interval:
return
try:
# 获取最近10分钟的数据与发言频率保持一致
recent_messages = message_api.get_messages_by_time_in_chat(
chat_id=self.chat_stream.stream_id,
start_time=current_time - 600, # 10分钟前
end_time=current_time,
filter_mai=True,
filter_command=True,
)
# 计算消息数量和用户数量
message_count = len(recent_messages)
user_ids = set()
for msg in recent_messages:
if msg.user_info and msg.user_info.user_id:
user_ids.add(msg.user_info.user_id)
user_count = len(user_ids)
# 获取当前小时的基准值
current_hour_base_messages, current_hour_base_users = self._get_current_hour_baseline()
# 计算当前小时平均每10分钟的基准值
current_hour_10min_messages = current_hour_base_messages / 6 # 1小时 = 6个10分钟
current_hour_10min_users = current_hour_base_users / 6
# 专注度调整逻辑:根据活跃度双向调整
# 检查是否有足够的数据进行分析
if user_count > 0 and current_hour_10min_users > 0 and message_count >= 2:
# 检查历史基准值是否有效(该时段有活跃度)
if current_hour_base_messages > 0.0 and current_hour_base_users > 0.0:
# 计算用户活跃度比率基于10分钟数据
user_ratio = user_count / current_hour_10min_users
# 计算消息活跃度比率基于10分钟数据
message_ratio = (
message_count / current_hour_10min_messages if current_hour_10min_messages > 0 else 1.0
)
# 双向调整逻辑
if user_ratio > 1.3 and message_ratio > 1.3:
# 活跃度很高提高专注度消耗更多LLM资源但回复更精准
target_focus_adjust = min(self.max_adjust, (user_ratio + message_ratio) / 2)
elif user_ratio > 1.1 and message_ratio > 1.1:
# 活跃度较高:适度提高专注度
target_focus_adjust = min(self.max_adjust, 1.0 + (user_ratio + message_ratio - 2.0) * 0.2)
elif user_ratio < 0.7 or message_ratio < 0.7:
# 活跃度很低降低专注度节省LLM资源
target_focus_adjust = max(self.min_adjust, min(user_ratio, message_ratio))
else:
# 正常情况:保持默认专注度
target_focus_adjust = 1.0
else:
# 历史基准值不足,不调整
target_focus_adjust = 1.0
else:
# 数据不足:不调整
target_focus_adjust = 1.0
# 限制调整范围
target_focus_adjust = max(self.min_adjust, min(self.max_adjust, target_focus_adjust))
# 记录调整前的值
old_focus_adjust = self.focus_value_adjust
# 平滑调整
self.focus_value_adjust = (
self.focus_value_adjust * (1 - self.smoothing_factor) + target_focus_adjust * self.smoothing_factor
)
# 计算当前小时平均每10分钟的基准值
current_hour_10min_messages = current_hour_base_messages / 6 # 1小时 = 6个10分钟
current_hour_10min_users = current_hour_base_users / 6
# 判断调整方向
if target_focus_adjust > 1.0:
adjust_direction = "提高"
elif target_focus_adjust < 1.0:
adjust_direction = "降低"
else:
if current_hour_base_messages <= 0.0 or current_hour_base_users <= 0.0:
adjust_direction = "不调整(该时段无活跃度)"
else:
adjust_direction = "保持"
# 计算实际变化方向
actual_change = ""
if self.focus_value_adjust > old_focus_adjust:
actual_change = f"{old_focus_adjust:.2f}x → {self.focus_value_adjust:.2f}x"
elif self.focus_value_adjust < old_focus_adjust:
actual_change = f"{old_focus_adjust:.2f}x → {self.focus_value_adjust:.2f}x"
else:
actual_change = f"无变化: {self.focus_value_adjust:.2f}x"
logger.info(
f"{self.log_prefix} 专注度调整: "
f"{user_count}名用户正在参与聊天,当前消息数: {message_count}|"
f"群基准: {current_hour_10min_messages:.2f}消息/{current_hour_10min_users:.2f}用户|"
f"[{adjust_direction}]{actual_change}"
)
except Exception as e:
logger.error(f"{self.log_prefix} 更新专注度调整值时出错: {e}")
def get_final_talk_frequency(self) -> float:
return (
get_config_base_talk_frequency(self.chat_stream.stream_id)
* self.get_dynamic_talk_frequency_adjust()
* self.talk_frequency_external_adjust
)
def get_final_focus_value(self) -> float:
return (
get_config_base_focus_value(self.chat_stream.stream_id)
* self.get_dynamic_focus_value_adjust()
* self.focus_value_external_adjust
)
def set_adjustment_parameters(
self,
min_adjust: Optional[float] = None,
max_adjust: Optional[float] = None,
base_message_count: Optional[int] = None,
base_user_count: Optional[int] = None,
smoothing_factor: Optional[float] = None,
update_interval: Optional[int] = None,
historical_update_interval: Optional[int] = None,
historical_days: Optional[int] = None,
):
"""
设置调整参数
Args:
min_adjust: 最小调整值
max_adjust: 最大调整值
base_message_count: 基准消息数量
base_user_count: 基准用户数量
smoothing_factor: 平滑因子
update_interval: 更新间隔(秒)
"""
if min_adjust is not None:
self.min_adjust = max(0.1, min_adjust)
if max_adjust is not None:
self.max_adjust = max(1.0, max_adjust)
if base_message_count is not None:
self.base_message_count = max(1, base_message_count)
if base_user_count is not None:
self.base_user_count = max(1, base_user_count)
if smoothing_factor is not None:
self.smoothing_factor = max(0.0, min(1.0, smoothing_factor))
if update_interval is not None:
self.update_interval = max(10, update_interval)
if historical_update_interval is not None:
self._historical_update_interval = max(300, historical_update_interval) # 最少5分钟
if historical_days is not None:
self._historical_days = max(1, min(30, historical_days)) # 1-30天之间
class FrequencyControlManager:
"""
频率控制管理器,管理多个聊天流的频率控制实例
"""
def __init__(self):
self.frequency_control_dict: Dict[str, FrequencyControl] = {}
def get_or_create_frequency_control(self, chat_id: str) -> FrequencyControl:
"""
获取或创建指定聊天流的频率控制实例
Args:
chat_id: 聊天流ID
Returns:
FrequencyControl: 频率控制实例
"""
if chat_id not in self.frequency_control_dict:
self.frequency_control_dict[chat_id] = FrequencyControl(chat_id)
return self.frequency_control_dict[chat_id]
# 创建全局实例
frequency_control_manager = FrequencyControlManager()