157 lines
5.0 KiB
Python
157 lines
5.0 KiB
Python
"""
|
|
train_single.py — 单品种训练入口
|
|
|
|
用法:
|
|
python train_single.py XAUUSD [--offline]
|
|
|
|
每个品种独立训练，checkpoint: checkpoints/ckpt_{symbol}_step_{N}.pt
|
|
训练完成后保存策略: strategies/best_{symbol}.json
|
|
"""
|
|
import sys, time, pathlib
|
|
from pathlib import Path
|
|
|
|
sys.path.insert(0, str(Path(__file__).parent))
|
|
|
|
from config import Config
|
|
from data_pipeline.fetcher import MT5DataFetcher
|
|
from data_pipeline.data_manager import MT5DataManager
|
|
from model_core.config import ModelConfig
|
|
from model_core.engine import AlphaEngine
|
|
|
|
|
|
def train_single(fetcher, symbol: str, offline: bool):
|
|
"""训练单个品种。
|
|
|
|
自动检测 checkpoint 续训，完成后保存策略到 strategies/best_{symbol}.json。
|
|
"""
|
|
import json, glob as _g
|
|
|
|
print(f"\n{'='*60}")
|
|
print(f" AlphaGPT 单品种训练 — {symbol}")
|
|
print(f"{'='*60}")
|
|
print(f" 奖励模式: {ModelConfig.REWARD_MODE}")
|
|
print(f" 训练步数: {ModelConfig.TRAIN_STEPS}")
|
|
print(f" K线目录: {Config.KLINE_CACHE_DIR}")
|
|
print(f" 离线模式: 是（仅读本地缓存）")
|
|
print(f"{'='*60}")
|
|
|
|
# 加载全部品种数据，然后切出单品种视图
|
|
original = Config.SYMBOLS[:]
|
|
Config.SYMBOLS = [symbol]
|
|
|
|
try:
|
|
mgr = MT5DataManager(fetcher)
|
|
mgr.load()
|
|
T = mgr.raw_dict["open"].shape[1]
|
|
print(f" 数据: {symbol} T={T} bars ({T/6240:.2f}年)")
|
|
except Exception as e:
|
|
print(f" [错误] 数据加载失败: {e}")
|
|
Config.SYMBOLS = original
|
|
return None
|
|
finally:
|
|
Config.SYMBOLS = original
|
|
|
|
engine = AlphaEngine(data_manager=mgr, target_symbol=symbol)
|
|
|
|
# 自动续训
|
|
ckpt_pattern = str(pathlib.Path("checkpoints") / f"ckpt_{symbol}_step_*.pt")
|
|
ckpt_files = sorted(_g.glob(ckpt_pattern))
|
|
start_step = 0
|
|
|
|
if ckpt_files:
|
|
latest = ckpt_files[-1]
|
|
try:
|
|
engine.load_checkpoint(latest)
|
|
start_step = engine._step if hasattr(engine, '_step') else int(
|
|
latest.split('_step_')[-1].replace('.pt', '')
|
|
)
|
|
print(f" [续训] 从 {latest} 恢复，start_step={start_step}")
|
|
except Exception as e:
|
|
print(f" [警告] checkpoint 加载失败: {e}，将从头开始")
|
|
|
|
if start_step >= ModelConfig.TRAIN_STEPS:
|
|
print(f" [完成] {symbol} 已完成全部 {ModelConfig.TRAIN_STEPS} 步，跳过训练")
|
|
_save_strategy(engine, symbol)
|
|
return engine
|
|
|
|
if start_step == 0:
|
|
print(f" [新训] 从 step 0 开始")
|
|
|
|
engine.train(start_step=start_step)
|
|
_save_strategy(engine, symbol)
|
|
return engine
|
|
|
|
|
|
def _save_strategy(engine, symbol):
|
|
"""保存单品种策略"""
|
|
import json
|
|
from model_core.vocab import VOCAB_VERSION
|
|
|
|
pathlib.Path("strategies").mkdir(exist_ok=True)
|
|
path = pathlib.Path("strategies") / f"best_{symbol}.json"
|
|
|
|
data = {
|
|
"vocab_version": VOCAB_VERSION,
|
|
"symbol": symbol,
|
|
"mode": "single",
|
|
"formula": engine.best_formula,
|
|
"formula_decoded": engine._decode_formula(engine.best_formula)
|
|
if engine.best_formula else None,
|
|
"best_score": engine.best_score,
|
|
"train_steps": ModelConfig.TRAIN_STEPS,
|
|
}
|
|
with open(path, "w") as f:
|
|
json.dump(data, f, indent=2, ensure_ascii=False)
|
|
print(f" 策略已保存: {path}")
|
|
|
|
|
|
# ── CLI ────────────────────────────────────────────────────────
|
|
if __name__ == "__main__":
|
|
offline = True
|
|
mode = "ftmo"
|
|
syms: list[str] = []
|
|
skip_next = False
|
|
for arg in sys.argv[1:]:
|
|
if skip_next:
|
|
skip_next = False
|
|
continue
|
|
if arg == "--mode":
|
|
skip_next = True
|
|
continue
|
|
if arg == "--cache-dir":
|
|
skip_next = True
|
|
continue
|
|
if arg.startswith("--"):
|
|
continue
|
|
syms.append(arg)
|
|
|
|
if "--mode" in sys.argv:
|
|
idx = sys.argv.index("--mode")
|
|
mode = sys.argv[idx + 1] if idx + 1 < len(sys.argv) else "ftmo"
|
|
if "--cache-dir" in sys.argv:
|
|
idx = sys.argv.index("--cache-dir")
|
|
if idx + 1 < len(sys.argv):
|
|
Config.KLINE_CACHE_DIR = sys.argv[idx + 1]
|
|
ModelConfig.REWARD_MODE = mode
|
|
|
|
if not syms:
|
|
print("用法: python train_single.py <SYMBOL> [--cache-dir PATH] [--mode ftmo]")
|
|
print(f"可选品种: {Config.TRAINABLE_SYMBOLS}")
|
|
sys.exit(1)
|
|
|
|
symbol = syms[0]
|
|
if symbol not in Config.TRAINABLE_SYMBOLS:
|
|
print(f"警告: {symbol} 不在 TRAINABLE_SYMBOLS 列表中，但仍将尝试训练")
|
|
|
|
t0 = time.time()
|
|
with MT5DataFetcher(offline=offline) as fetcher:
|
|
eng = train_single(fetcher, symbol, offline)
|
|
|
|
elapsed = time.time() - t0
|
|
if eng:
|
|
print(f"\n<<< [{symbol}] 完成: score={eng.best_score:.4f} 耗时 {elapsed/3600:.2f}h")
|
|
if eng.best_formula:
|
|
print(f" {eng._decode_formula(eng.best_formula)}")
|
|
else:
|
|
print(f"\n<<< [{symbol}] 失败")
|