feat(examples): add US stock day prediction script

examples/predict_us_stock_day.py

@@ -0,0 +1,101 @@
+import argparse
+import os
+from datetime import timedelta
+
+import pandas as pd
+import yfinance as yf
+import torch
+
+import sys
+sys.path.append("../")
+from model import Kronos, KronosTokenizer, KronosPredictor
+
+
+def pick_device() -> str:
+    if torch.cuda.is_available():
+        return "cuda:0"
+    if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+        return "mps"
+    return "cpu"
+
+
+def load_us_daily(symbol: str, min_rows: int = 520) -> pd.DataFrame:
+    # Pull enough daily candles to satisfy lookback while preserving recent context.
+    df = yf.download(symbol, period="5y", interval="1d", auto_adjust=False, progress=False)
+    if df is None or df.empty:
+        raise RuntimeError(f"No data returned for {symbol} from Yahoo Finance")
+
+    df = df.reset_index().rename(
+        columns={
+            "Date": "timestamps",
+            "Open": "open",
+            "High": "high",
+            "Low": "low",
+            "Close": "close",
+            "Volume": "volume",
+        }
+    )
+    df["timestamps"] = pd.to_datetime(df["timestamps"]).dt.tz_localize(None)
+    df["amount"] = df["close"] * df["volume"]
+    df = df[["timestamps", "open", "high", "low", "close", "volume", "amount"]]
+    df = df.dropna().sort_values("timestamps").reset_index(drop=True)
+
+    if len(df) < min_rows:
+        raise RuntimeError(
+            f"Not enough rows for robust context. Need >= {min_rows}, got {len(df)} for {symbol}."
+        )
+
+    return df
+
+
+def future_business_days(last_day: pd.Timestamp, periods: int) -> pd.Series:
+    start = last_day + timedelta(days=1)
+    return pd.Series(pd.bdate_range(start=start, periods=periods))
+
+
+def main(symbol: str, pred_len: int, lookback: int, out_dir: str) -> None:
+    device = pick_device()
+    print(f"Using device: {device}")
+
+    tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
+    model = Kronos.from_pretrained("NeoQuasar/Kronos-small")
+    predictor = KronosPredictor(model, tokenizer, device=device, max_context=512)
+
+    df = load_us_daily(symbol)
+    hist = df.tail(lookback).copy()
+
+    x_df = hist[["open", "high", "low", "close", "volume", "amount"]]
+    x_timestamp = hist["timestamps"]
+    y_timestamp = future_business_days(hist["timestamps"].iloc[-1], pred_len)
+
+    pred_df = predictor.predict(
+        df=x_df,
+        x_timestamp=x_timestamp,
+        y_timestamp=y_timestamp,
+        pred_len=pred_len,
+        T=1.0,
+        top_p=0.9,
+        sample_count=1,
+        verbose=False,
+    ).reset_index().rename(columns={"index": "timestamps"})
+
+    os.makedirs(out_dir, exist_ok=True)
+    out_file = os.path.join(out_dir, f"pred_{symbol}_{pred_len}d.csv")
+    pred_df.to_csv(out_file, index=False)
+
+    print("\nLast observed row:")
+    print(hist.tail(1).to_string(index=False))
+    print("\nForecast:")
+    print(pred_df[["timestamps", "open", "high", "low", "close", "volume", "amount"]].to_string(index=False))
+    print(f"\nSaved: {out_file}")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Predict next N business days for a US stock with Kronos")
+    parser.add_argument("--symbol", type=str, default="F", help="Ticker symbol, e.g. F")
+    parser.add_argument("--pred-len", type=int, default=3, help="Number of future business days")
+    parser.add_argument("--lookback", type=int, default=400, help="Historical lookback rows")
+    parser.add_argument("--out-dir", type=str, default="./outputs", help="Output directory")
+    args = parser.parse_args()
+
+    main(symbol=args.symbol, pred_len=args.pred_len, lookback=args.lookback, out_dir=args.out_dir)