feat(proxy-clients): pydantic-ai / mem0 / graphiti integration layer

.claude/launch.json

@@ -0,0 +1,9 @@
+{
+  "version": "0.0.1",
+  "configurations": [
+    {
+      "name": "stock-tracker",
+      "url": "http://localhost:8080"
+    }
+  ]
+}

outputs/data/grid.json

@@ -0,0 +1,160 @@
+{
+  "grid": [
+    [
+      3,
+      20
+    ],
+    [
+      3,
+      30
+    ],
+    [
+      3,
+      40
+    ],
+    [
+      3,
+      60
+    ],
+    [
+      3,
+      90
+    ],
+    [
+      3,
+      120
+    ],
+    [
+      5,
+      20
+    ],
+    [
+      5,
+      30
+    ],
+    [
+      5,
+      40
+    ],
+    [
+      5,
+      60
+    ],
+    [
+      5,
+      90
+    ],
+    [
+      5,
+      120
+    ],
+    [
+      8,
+      20
+    ],
+    [
+      8,
+      30
+    ],
+    [
+      8,
+      40
+    ],
+    [
+      8,
+      60
+    ],
+    [
+      8,
+      90
+    ],
+    [
+      8,
+      120
+    ],
+    [
+      10,
+      20
+    ],
+    [
+      10,
+      30
+    ],
+    [
+      10,
+      40
+    ],
+    [
+      10,
+      60
+    ],
+    [
+      10,
+      90
+    ],
+    [
+      10,
+      120
+    ],
+    [
+      15,
+      20
+    ],
+    [
+      15,
+      30
+    ],
+    [
+      15,
+      40
+    ],
+    [
+      15,
+      60
+    ],
+    [
+      15,
+      90
+    ],
+    [
+      15,
+      120
+    ],
+    [
+      20,
+      30
+    ],
+    [
+      20,
+      40
+    ],
+    [
+      20,
+      60
+    ],
+    [
+      20,
+      90
+    ],
+    [
+      20,
+      120
+    ]
+  ],
+  "fast_pool": [
+    3,
+    5,
+    8,
+    10,
+    15,
+    20
+  ],
+  "slow_pool": [
+    20,
+    30,
+    40,
+    60,
+    90,
+    120
+  ]
+}

outputs/scripts/backtest_grid.py

@@ -0,0 +1,156 @@
+"""批量向量化回测金叉死叉网格，输出 ma_grid_results.csv。
+
+策略：
+- 金叉（fast MA 上穿 slow MA）建仓，死叉（fast MA 下穿 slow MA）平仓
+- 前复权数据回测（A 股标准做法）
+
+A 股真实性优化（v2）：
+- T+1：信号 t 日收盘产生，t+1 日才成交（位移 1 bar）
+- 成交价：次日开盘价 open（非收盘价，无未来函数）
+- 成本：佣金万1.25(单边)+印花税千1(仅卖出)+过户费万0.1 ≈ 单边千0.65
+  → vectorbt fees 按双边统一收，取 fees=0.00065, slippage=0.0005（偏保守）
+- 涨跌停：000001 大盘银行股 5年仅2天近涨停，影响极小，未加过滤
+
+输出：ma_grid_results.csv（含原始 v1 与优化 v2 双列对比）
+"""
+
+import json
+from pathlib import Path
+
+import pandas as pd
+import vectorbt as vbt
+
+DATA_DIR = Path(__file__).resolve().parent.parent / "data"
+OUT_DIR = Path(__file__).resolve().parent.parent / "reports"
+OUT_DIR.mkdir(parents=True, exist_ok=True)
+
+DATA_FILE = DATA_DIR / "000001_front.parquet"
+GRID_FILE = DATA_DIR / "grid.json"
+RESULT_FILE = OUT_DIR / "ma_grid_results.csv"
+
+INIT_CASH = 100_000
+FREQ = "D"
+# A 股真实成本：佣金万2.5(双边)+印花税千1(单边卖出)+过户费万0.1
+# 单边综合约 千0.65；vectorbt fees 作用于每笔成交(买卖各收一次)
+FEES = 0.00065
+SLIPPAGE = 0.0005
+
+
+def load_data():
+    df = pd.read_parquet(DATA_FILE)
+    df["date"] = pd.to_datetime(df["date"])
+    df = df.set_index("date").sort_index()
+    return df["close"].astype(float), df["open"].astype(float)
+
+
+def run_grid(close, open_px, grid):
+    rows = []
+    # 基准：买入持有（次日开盘全仓）
+    bh_pf = vbt.Portfolio.from_holding(close, init_cash=INIT_CASH, freq=FREQ)
+    bh_total = float(bh_pf.total_return())
+    bh_sharpe = float(bh_pf.sharpe_ratio())
+    bh_maxdd = float(bh_pf.max_drawdown())
+    print(f"[基准 买入持有] 总收益={bh_total:.4f} 夏普={bh_sharpe:.3f} 最大回撤={bh_maxdd:.4f}")
+
+    for fast, slow in grid:
+        fast_ma = vbt.MA.run(close, fast).ma
+        slow_ma = vbt.MA.run(close, slow).ma
+        entries = fast_ma.vbt.crossed_above(slow_ma)
+        exits = fast_ma.vbt.crossed_below(slow_ma)
+
+        # v1: 原始（收盘信号+收盘成交，允许T+0）
+        pf_v1 = vbt.Portfolio.from_signals(
+            close,
+            entries=entries,
+            exits=exits,
+            init_cash=INIT_CASH,
+            freq=FREQ,
+            fees=0.001,
+            slippage=0.001,
+        )
+
+        # v2: A股真实（信号位移1bar → 次日成交，T+1）
+        # pandas shift(1) 把信号往后挪一天：t日信号→t+1日成交（无未来函数）
+        # 注意：不能用 vbt.signals.bshift，它往前挪=未来函数
+        # fillna 后强制 bool，避免 numba 报 object 类型错
+        entries_s = entries.shift(1).fillna(False).astype(bool)
+        exits_s = exits.shift(1).fillna(False).astype(bool)
+        pf_v2 = vbt.Portfolio.from_signals(
+            close,
+            entries=entries_s,
+            exits=exits_s,
+            price=open_px,  # 次日开盘价成交
+            init_cash=INIT_CASH,
+            freq=FREQ,
+            fees=FEES,
+            slippage=SLIPPAGE,
+        )
+
+        def metrics(pf):
+            trades = pf.trades.records_readable
+            n = len(trades)
+            wr = float((trades["PnL"] > 0).mean()) if n > 0 else 0.0
+            return {
+                "total_return": float(pf.total_return()),
+                "annual_return": float(pf.annualized_return()),
+                "sharpe": float(pf.sharpe_ratio()),
+                "max_drawdown": float(pf.max_drawdown()),
+                "n_trades": n,
+                "win_rate": wr,
+            }
+
+        m1, m2 = metrics(pf_v1), metrics(pf_v2)
+        rows.append(
+            {
+                "fast": fast,
+                "slow": slow,
+                **{f"{k}_v1": m1[k] for k in m1},
+                **{f"{k}_v2": m2[k] for k in m2},
+            }
+        )
+
+    res = pd.DataFrame(rows)
+    res["bh_total_return"] = bh_total
+    res["bh_sharpe"] = bh_sharpe
+    res["bh_max_drawdown"] = bh_maxdd
+    res["excess_return_v1"] = res["total_return_v1"] - bh_total
+    res["excess_return_v2"] = res["total_return_v2"] - bh_total
+    return res, {"bh_total_return": bh_total, "bh_sharpe": bh_sharpe, "bh_max_drawdown": bh_maxdd}
+
+
+if __name__ == "__main__":
+    close, open_px = load_data()
+    print(
+        f"数据: {close.shape[0]} 个交易日  {close.index.min().date()} ~ {close.index.max().date()}"
+    )
+    grid = json.loads(GRID_FILE.read_text(encoding="utf-8"))["grid"]
+    print(f"网格: {len(grid)} 组  (v1原始 vs v2 T+1+次日开盘)")
+    res, bh = run_grid(close, open_px, grid)
+    res.to_csv(RESULT_FILE, index=False, encoding="utf-8-sig")
+    print(f"\n=== 结果已存 {RESULT_FILE} ===")
+    print("\n--- v2 优化版 Top5 (按 total_return_v2) ---")
+    print(
+        res.sort_values("total_return_v2", ascending=False)
+        .head(5)[
+            [
+                "fast",
+                "slow",
+                "total_return_v1",
+                "total_return_v2",
+                "sharpe_v1",
+                "sharpe_v2",
+                "n_trades_v1",
+                "n_trades_v2",
+            ]
+        ]
+        .to_string(index=False)
+    )
+    print(f"\n基准买入持有: 总收益={bh['bh_total_return']:.4f} 夏普={bh['bh_sharpe']:.3f}")
+    # v1 vs v2 整体对比
+    print(f"\n--- v1 vs v2 整体对比 ---")
+    print(
+        f"v1 跑赢基准: {int((res['excess_return_v1'] > 0).sum())}/{len(res)}  v2 跑赢基准: {int((res['excess_return_v2'] > 0).sum())}/{len(res)}"
+    )
+    print(
+        f"v1 平均交易次数: {res['n_trades_v1'].mean():.1f}  v2 平均交易次数: {res['n_trades_v2'].mean():.1f}  (T+1应使v2略低)"
+    )

outputs/scripts/design_grid.py

@@ -0,0 +1,33 @@
+"""金叉死叉策略参数网格设计。
+
+规则：
+- 短周期 fast ∈ [3,5,8,10,15,20]，长周期 slow ∈ [20,30,40,60,90,120]
+- 有效组合要求 fast < slow（否则均线无交叉意义）
+- 金叉（fast 上穿 slow）做多，死叉（fast 下穿 slow）平仓
+- 全部用前复权数据回测（A 股回测标准做法）
+"""
+
+import json
+from pathlib import Path
+
+FAST = [3, 5, 8, 10, 15, 20]
+SLOW = [20, 30, 40, 60, 90, 120]
+
+# 剔除 fast >= slow 的无效组合
+GRID = [(f, s) for f in FAST for s in SLOW if f < s]
+
+OUT = Path(__file__).resolve().parent.parent / "data" / "grid.json"
+
+if __name__ == "__main__":
+    OUT.parent.mkdir(parents=True, exist_ok=True)
+    OUT.write_text(
+        json.dumps(
+            {"grid": GRID, "fast_pool": FAST, "slow_pool": SLOW}, ensure_ascii=False, indent=2
+        ),
+        encoding="utf-8",
+    )
+    print(f"有效组合数: {len(GRID)}")
+    print(f"fast 候选: {FAST}")
+    print(f"slow 候选: {SLOW}")
+    print(f"网格: {GRID}")
+    print(f"-> {OUT}")