feat: 增加早期 FSRS 支持

This commit is contained in:
2026-04-25 01:54:18 +08:00
parent 7f2bcf45de
commit fa2f8fa701
5 changed files with 247 additions and 4 deletions

View File

@@ -2,17 +2,20 @@ from .base import BaseAlgorithm
from .sm2 import SM2Algorithm
from .sm15m import SM15MAlgorithm
from .nsp0 import NSP0Algorithm
from .fsrs import FSRSAlgorithm
__all__ = [
"SM2Algorithm",
"BaseAlgorithm",
"SM15MAlgorithm",
"NSP0Algorithm",
"FSRSAlgorithm",
]
algorithms = {
"SM-2": SM2Algorithm,
"NSP-0": NSP0Algorithm,
"SM-15M": SM15MAlgorithm,
"FSRS": FSRSAlgorithm,
"Base": BaseAlgorithm,
}

View File

@@ -1,6 +1,246 @@
# FSRS 算法模块, 尚未就绪
"""
FSRS 算法模块 — 基于 py-fsrs 的现代间隔重复调度器
基于: https://github.com/open-spaced-repetition/py-fsrs
"""
import json
import os
import pathlib
from datetime import datetime, timezone, timedelta
from typing import TypedDict
from fsrs import Scheduler, Card, Rating
from heurams.context import config_var
from heurams.services.logger import get_logger
from heurams.services.timer import get_daystamp, get_timestamp
from .base import BaseAlgorithm
logger = get_logger(__name__)
logger.info("FSRS算法模块尚未实现")
# 全局 Scheduler 状态文件路径
_SCHEDULER_STATE_FILE = pathlib.Path(
config_var.get()["global"]["paths"]["data"]
) / "global" / "fsrs_scheduler_state.json"
def _get_global_scheduler():
"""获取全局 FSRS Scheduler 实例,从文件加载或创建新的"""
if os.path.exists(_SCHEDULER_STATE_FILE):
try:
with open(_SCHEDULER_STATE_FILE, "r", encoding="utf-8") as f:
return Scheduler.from_json(f.read())
except Exception:
logger.warning("FSRS Scheduler 状态文件加载失败,创建新实例")
return Scheduler()
def _save_global_scheduler(scheduler):
"""保存全局 FSRS Scheduler 实例到文件"""
try:
_SCHEDULER_STATE_FILE.parent.mkdir(parents=True, exist_ok=True)
data = scheduler.to_json()
with open(_SCHEDULER_STATE_FILE, "w", encoding="utf-8") as f:
f.write(data)
except Exception:
logger.exception("FSRS Scheduler 状态保存失败")
def _feedback_to_rating(feedback: int) -> Rating:
"""将 SM-2 风格 feedback (0-5) 映射为 FSRS Rating (1-4)"""
if feedback <= 2:
return Rating.Again
elif feedback == 3:
return Rating.Hard
elif feedback == 4:
return Rating.Good
else:
return Rating.Easy
def _datetime_to_daystamp(dt: datetime) -> int:
"""将 datetime 转换为天数戳(从 1970-01-01"""
epoch = datetime(1970, 1, 1, tzinfo=timezone.utc)
delta = dt - epoch
return delta.days
def _daystamp_to_datetime(daystamp: int) -> datetime:
"""将天数戳转换为 UTC datetime"""
epoch = datetime(1970, 1, 1, tzinfo=timezone.utc)
return epoch + timedelta(days=daystamp)
class FSRSAlgorithm(BaseAlgorithm):
algo_name = "FSRS"
desc = "基于 py-fsrs 的现代间隔重复调度器"
class AlgodataDict(TypedDict):
# FSRS 特有字段
fsrs_state: int # State 枚举值: 1=Learning, 2=Review, 3=Relearning
fsrs_step: int # 当前学习步进索引, -1 表示 None (Review 状态)
fsrs_stability: float # 稳定性0.0 表示尚未计算
fsrs_difficulty: float # 难度 [1.0, 10.0], 0.0 表示尚未计算
# 标准 BaseAlgorithm 兼容字段
real_rept: int
rept: int
interval: int
last_date: int
next_date: int
is_activated: int
last_modify: float
defaults = {
"fsrs_state": 1, # State.Learning
"fsrs_step": 0,
"fsrs_stability": 0.0,
"fsrs_difficulty": 0.0,
"real_rept": 0,
"rept": 0,
"interval": 0,
"last_date": 0,
"next_date": 0,
"is_activated": 0,
"last_modify": get_timestamp(),
}
@classmethod
def _algodata_to_card(cls, algodata):
"""从 algodata 恢复 Card 实例"""
data = algodata.get(cls.algo_name, cls.defaults.copy())
card = Card()
# State: int → IntEnum
card.state = data.get("fsrs_state", 1)
# Step: -1 表示 NoneReview 状态下的 card.step 为 None
step = data.get("fsrs_step", -1)
card.step = None if step == -1 else step
# Stability: 0.0 表示尚未计算(新卡片)
stability = data.get("fsrs_stability", 0.0)
card.stability = None if stability == 0.0 else stability
# Difficulty: 0.0 表示尚未计算
difficulty = data.get("fsrs_difficulty", 0.0)
card.difficulty = None if difficulty == 0.0 else difficulty
# due: 新卡片(next_date ≤ 0)设为当前时间
next_date = data.get("next_date", 0)
if next_date <= 0:
card.due = datetime.now(timezone.utc)
else:
card.due = _daystamp_to_datetime(next_date)
# last_review
last_date = data.get("last_date", 0)
card.last_review = (
_daystamp_to_datetime(last_date) if last_date > 0 else None
)
return card
@classmethod
def _card_to_algodata(cls, card, algodata):
"""将 Card 实例状态写回 algodata"""
if cls.algo_name not in algodata:
algodata[cls.algo_name] = cls.defaults.copy()
data = algodata[cls.algo_name]
data["fsrs_state"] = int(card.state)
data["fsrs_step"] = card.step if card.step is not None else -1
data["fsrs_stability"] = card.stability if card.stability is not None else 0.0
data["fsrs_difficulty"] = (
card.difficulty if card.difficulty is not None else 0.0
)
data["last_date"] = (
_datetime_to_daystamp(card.last_review)
if card.last_review
else data.get("last_date", 0)
)
data["next_date"] = (
_datetime_to_daystamp(card.due) if card.due else 0
)
data["interval"] = max(0, data["next_date"] - data["last_date"])
data["last_modify"] = get_timestamp()
return algodata
@classmethod
def revisor(
cls, algodata: dict, feedback: int = 5, is_new_activation: bool = False
):
"""FSRS 算法迭代决策机制实现
将 feedback (0-5) 映射为 FSRS Rating 后交由 py-fsrs 调度器处理。
Args:
feedback (int): 0-5 的记忆保留率量化参数
is_new_activation: 是否为全新激活(重置为初始状态)
"""
logger.debug(
"FSRS.revisor 开始, feedback: %d, is_new_activation: %s",
feedback,
is_new_activation,
)
if feedback == -1:
logger.debug("feedback 为 -1, 跳过更新")
return
scheduler = _get_global_scheduler()
rating = _feedback_to_rating(feedback)
if is_new_activation:
card = Card()
logger.debug("新激活,创建新 Card")
else:
card = cls._algodata_to_card(algodata)
card, review_log = scheduler.review_card(card, rating)
_save_global_scheduler(scheduler)
cls._card_to_algodata(card, algodata)
# real_rept: 总复习次数
algodata[cls.algo_name]["real_rept"] += 1
# rept: 成功回忆次数feedback ≥ 3 视为成功)
if feedback >= 3:
algodata[cls.algo_name]["rept"] += 1
logger.debug(
"FSRS.revisor 完成: stability=%s, difficulty=%s, state=%s, "
"next_date=%d",
card.stability,
card.difficulty,
card.state,
algodata[cls.algo_name]["next_date"],
)
@classmethod
def is_due(cls, algodata):
data = algodata.get(cls.algo_name, cls.defaults.copy())
next_date = data.get("next_date", 0)
current = get_daystamp()
result = next_date <= current
logger.debug(
"FSRS.is_due: next_date=%d, current=%d, result=%s",
next_date,
current,
result,
)
return result
@classmethod
def get_rating(cls, algodata):
data = algodata.get(cls.algo_name, cls.defaults.copy())
difficulty = data.get("fsrs_difficulty", 0.0)
logger.debug("FSRS.get_rating: difficulty=%f", difficulty)
return str(difficulty)
@classmethod
def nextdate(cls, algodata) -> int:
data = algodata.get(cls.algo_name, cls.defaults.copy())
next_date = data.get("next_date", 0)
logger.debug("FSRS.nextdate: %d", next_date)
return next_date