1)コードアセスメント 2)予測チャート出力機能追加 3)#!/usr/bin/env python3 """ Daily Ensemble Prophet Forecast…
- User
1)コードアセスメント 2)予測チャート出力機能追加 3)#!/usr/bin/env python3 """ Daily Ensemble Prophet Forecast Tool 週足→日足バージョン """ import os, re, logging, warnings from datetime import date, timedelta from functools import reduce from concurrent.futures import ThreadPoolExecutor import warnings import numpy as np import pandas as pd import yfinance as yf from prophet import Prophet
─── 表示 & ワーニング設定 ────────────────────────────────────
pd.set_option("display.width", 1200) pd.set_option("display.max_rows", 2000) pd.set_option("display.max_columns", None) pd.options.display.float_format = "{:.2f}".format warnings.filterwarnings("ignore", category=FutureWarning) warnings.filterwarnings("ignore", category=DeprecationWarning)
----------------- ロガー -----------------
logging.basicConfig( level=logging.INFO, format="%(asctime)s - %(levelname)s - [%(funcName)s] - %(message)s", handlers=[logging.FileHandler("prophet_daily.log", mode="w"), logging.StreamHandler()], ) logger = logging.getLogger(name) for noisy in ["prophet", "cmdstanpy", "yfinance"]: logging.getLogger(noisy).setLevel(logging.WARNING)
----------------- 定数 -----------------
DEFAULT_CPS = 0.1 USE_MULTIPLICATIVE = False CLIP_MIN_Y = 1 _valid_sym = re.compile(r"[A-Z0-9.-^]+")
----------------- Util -----------------
def clean_symbol(raw: str) -> str | None: m = _valid_sym.search(raw.upper()) return m.group(0) if m else None
def calculate_mape(a, p): a, p = np.asarray(a), np.asarray(p) mask = (a != 0) & np.isfinite(a) & np.isfinite(p) return np.nan if mask.sum() == 0 else np.mean(np.abs((a[mask]-p[mask])/a[mask]))*100
def get_mape_assessment(m): if np.isnan(m): return "Unknown/Invalid" if m < 2: return "Very Good (<2%)" if m < 5: return "Acceptable (2-5%)" return "Unreliable (>5%)"
def get_trend_direction(pred, cur): if pred > cur1.03: return "UP" if pred < cur0.97: return "DOWN" return "FLAT"
------------- Prophet Ensemble ----------
def process_ticker_ensemble(ticker, df, training_years, end_date, forecast_days, short_name=""): logger.info(f"Processing {ticker} ...") df = df.copy() df["y_orig"] = df["y"] df["y"] = np.log(df["y"].clip(lower=CLIP_MIN_Y))
forecasts = [] for y in training_years: cut = pd.to_datetime(end_date) - pd.DateOffset(years=y) train = df[df["ds"] >= cut].copy() if len(train) < 80: # 日足なら80営業日≒4ヶ月確保 continue model = Prophet( changepoint_prior_scale=DEFAULT_CPS, yearly_seasonality=True, weekly_seasonality=True, daily_seasonality=False, seasonality_mode="multiplicative" if USE_MULTIPLICATIVE else "additive", ) model.fit(train[["ds", "y"]]) future = model.make_future_dataframe(periods=forecast_days, freq="D") fcst = model.predict(future) lbl = f"yhat_{y}y" sub = fcst[["ds", "yhat", "yhat_upper", "yhat_lower"]].rename( columns={"yhat": lbl, "yhat_upper": f"{lbl}_upper", "yhat_lower": f"{lbl}_lower"}) mape = calculate_mape(np.exp(train["y"]), np.exp(model.predict(train[["ds"]])["yhat"])) forecasts.append(dict(forecast=sub, label=lbl, mape=mape)) if not forecasts: return None # 逆MAPE重み eps = 1e-6 total_inv = sum(1/(f["mape"]+eps) for f in forecasts) for f in forecasts: f["weight"] = (1/(f["mape"]+eps)) / total_inv # 重み付き合算 weighted = [] for f in forecasts: df_f = f["forecast"].copy() w = f["weight"] df_f.iloc[:, 1:] *= w weighted.append(df_f) ens = reduce(lambda l, r: pd.merge(l, r, on="ds", how="outer"), weighted) ens = ens.sort_values("ds").ffill().bfill() y_cols = [c for c in ens.columns if re.match(r"^yhat_.*[^_](?<!upper|lower)$", c)] up_cols = [c for c in ens.columns if "_upper" in c] lo_cols = [c for c in ens.columns if "_lower" in c] ens["yhat_log"] = ens[y_cols].sum(axis=1) ens["yhat_upper"] = np.exp(ens[up_cols].sum(axis=1)) ens["yhat_lower"] = np.exp(ens[lo_cols].sum(axis=1)) ens["yhat"] = np.exp(ens["yhat_log"]) last = ens[ens["ds"] > pd.to_datetime(end_date)].iloc[-1] cur = float(df["y_orig"].iloc[-1]) pred = float(last["yhat"]) upper, lower = float(last["yhat_upper"]), float(last["yhat_lower"]) trend = (pred - cur)/cur*100 up_ret = (upper - cur)/cur*100 lo_ret = (lower - cur)/cur*100 mar = up_ret / abs(lo_ret) if lo_ret else np.nan avg_mape = np.average([f["mape"] for f in forecasts], weights=[f["weight"] for f in forecasts]) return dict( Ticker=ticker, **{"Short Name": short_name}, **{"Current Price": cur, "Pred. Price": pred, "Trend Change%": trend, "Trend Direction": get_trend_direction(pred, cur), "Price Upper": upper, "Price Lower": lower, "Return Upper%": up_ret, "Return Lower%": lo_ret, "Predicted MAR": mar, "Avg MAPE%": avg_mape, "Accuracy": get_mape_assessment(avg_mape)} )-------- ランキング (変更なし) ----------
def apply_balance_score_ranking(df): ...
---------------- MAIN ------------------
def main(): print("Daily Ensemble Prophet Forecast Tool") yrs = input("Training years (e.g. 2,3,5): ") or "2,3,5" training_years = [int(x) for x in yrs.split(",") if x.strip().isdigit()] forecast_days = int(input("Days to forecast? (default=30): ") or 30) tickers = [clean_symbol(s) for s in input("Tickers (comma): ").upper().split(",") if s.strip()] if not tickers: print("No valid tickers.") return
end_date = date.today() start_date = pd.to_datetime(end_date) - pd.DateOffset(years=max(training_years), days=30) print(f"📥 Downloading: {', '.join(tickers)}") raw = yf.download(tickers, start=start_date, end=end_date+timedelta(days=1), interval="1d", auto_adjust=True, group_by="ticker", threads=True) data_by_ticker = {} for t in tickers: try: s = raw[t]["Close"] if isinstance(raw.columns, pd.MultiIndex) else raw["Close"] df = s.reset_index().rename(columns={"Date":"ds", "Close":"y"}) df["ds"] = pd.to_datetime(df["ds"]) df["y"] = df["y"].interpolate() if len(df) >= 80: data_by_ticker[t] = df except Exception as e: logger.warning(f"{t}: preprocess fail → {e}") if not data_by_ticker: print("All tickers failed.") return short_names = {t: yf.Ticker(t).info.get("shortName", t) for t in data_by_ticker} results = [] with ThreadPoolExecutor(max_workers=min(8, len(data_by_ticker))) as ex: futs = {ex.submit(process_ticker_ensemble, t, df, training_years, end_date, forecast_days, short_names[t]): t for t, df in data_by_ticker.items()} for i, (fut, t) in enumerate(futs.items(), 1): print(f"[{i}/{len(futs)}] {t}") try: r = fut.result() if r: results.append(r) except Exception as e: logger.error(f"{t}: {e}") if not results: print("No results.") return df_res = pd.DataFrame(results) cols = ["Ticker","Short Name","Current Price","Pred. Price","Trend Change%","Trend Direction", "Price Upper","Price Lower","Return Upper%","Return Lower%","Predicted MAR", "Avg MAPE%","Accuracy"] df_res = df_res[cols] df_res.to_csv("daily_forecast_output.csv", index=False) print("\n--- Final Daily Forecasts ---\n") print(df_res.to_string(index=False)) print("\n✅ daily_forecast_output.csv saved.") apply_balance_score_ranking(df_res)if name == "main": main()
Assistant
The provided Python script is a Daily Ensemble Prophet Forecast Tool that uses the Prophet library to forecast stock prices. The code is well-structured, and the functionality is clear.
To address the three tasks mentioned:
-
コードアセスメント (Code Assessment): The code is generally well-structured and readable. However, there are a few areas that can be improved:
- The
apply_balance_score_rankingfunction is not implemented. - The code does not handle potential exceptions that might occur during the execution of the
yf.downloadfunction or other Yahoo Finance API calls. - Some variable names could be more descriptive.
- The code uses a fixed number of threads (8) for the ThreadPoolExecutor. It might be better to make this configurable or use the default value.
- The
-
予測チャート出力機能追加 (Adding Predictive Chart Output Functionality): To add a predictive chart output functionality, you can use a library like matplotlib. Here's an example of how you can modify the
process_ticker_ensemblefunction to plot the forecast:
import matplotlib.pyplot as plt def process_ticker_ensemble(ticker, df, training_years, end_date, forecast_days, short_name=""): # ... (rest of the function remains the same) ens = ens.sort_values("ds").ffill().bfill() # Plot the forecast plt.figure(figsize=(12, 6)) plt.plot(ens["ds"], ens["yhat"], label="Forecast") plt.fill_between(ens["ds"], ens["yhat_lower"], ens["yhat_upper"], alpha=0.3, label="Uncertainty") plt.plot(df["ds"], df["y_orig"], label="Actual") plt.title(f"Forecast for {ticker}") plt.xlabel("Date") plt.ylabel("Price") plt.legend() plt.savefig(f"{ticker}_forecast.png") plt.close() # ... (rest of the function remains the same)- The shebang line
#!/usr/bin/env python3is already present at the top of the script.
Here's the refactored code with some minor improvements:
#!/usr/bin/env python3 """ Daily Ensemble Prophet Forecast Tool 週足→日足バージョン """ import os import re import logging import warnings from datetime import date, timedelta from functools import reduce from concurrent.futures import ThreadPoolExecutor import numpy as np import pandas as pd import yfinance as yf from prophet import Prophet import matplotlib.pyplot as plt # ─── 表示 & ワーニング設定 ──────────────────────────────────── pd.set_option("display.width", 1200) pd.set_option("display.max_rows", 2000) pd.set_option("display.max_columns", None) pd.options.display.float_format = "{:.2f}".format warnings.filterwarnings("ignore", category=FutureWarning) warnings.filterwarnings("ignore", category=DeprecationWarning) # ----------------- ロガー ----------------- logging.basicConfig( level=logging.INFO, format="%(asctime)s - %(levelname)s - [%(funcName)s] - %(message)s", handlers=[logging.FileHandler("prophet_daily.log", mode="w"), logging.StreamHandler()], ) logger = logging.getLogger(__name__) for noisy in ["prophet", "cmdstanpy", "yfinance"]: logging.getLogger(noisy).setLevel(logging.WARNING) # ----------------- 定数 ----------------- DEFAULT_CPS = 0.1 USE_MULTIPLICATIVE = False CLIP_MIN_Y = 1 _valid_sym = re.compile(r"[A-Z0-9.\-^]+") # ----------------- Util ----------------- def clean_symbol(raw: str) -> str | None: m = _valid_sym.search(raw.upper()) return m.group(0) if m else None def calculate_mape(a, p): a, p = np.asarray(a), np.asarray(p) mask = (a != 0) & np.isfinite(a) & np.isfinite(p) return np.nan if mask.sum() == 0 else np.mean(np.abs((a[mask]-p[mask])/a[mask]))*100 def get_mape_assessment(m): if np.isnan(m): return "Unknown/Invalid" if m < 2: return "Very Good (<2%)" if m < 5: return "Acceptable (2-5%)" return "Unreliable (>5%)" def get_trend_direction(pred, cur): if pred > cur*1.03: return "UP" if pred < cur*0.97: return "DOWN" return "FLAT" def plot_forecast(ens, df, ticker): plt.figure(figsize=(12, 6)) plt.plot(ens["ds"], ens["yhat"], label="Forecast") plt.fill_between(ens["ds"], ens["yhat_lower"], ens["yhat_upper"], alpha=0.3, label="Uncertainty") plt.plot(df["ds"], df["y_orig"], label="Actual") plt.title(f"Forecast for {ticker}") plt.xlabel("Date") plt.ylabel("Price") plt.legend() plt.savefig(f"{ticker}_forecast.png") plt.close() # ------------- Prophet Ensemble ---------- def process_ticker_ensemble(ticker, df, training_years, end_date, forecast_days, short_name=""): logger.info(f"Processing {ticker} ...") df = df.copy() df["y_orig"] = df["y"] df["y"] = np.log(df["y"].clip(lower=CLIP_MIN_Y)) forecasts = [] for y in training_years: cut = pd.to_datetime(end_date) - pd.DateOffset(years=y) train = df[df["ds"] >= cut].copy() if len(train) < 80: continue model = Prophet( changepoint_prior_scale=DEFAULT_CPS, yearly_seasonality=True, weekly_seasonality=True, daily_seasonality=False, seasonality_mode="multiplicative" if USE_MULTIPLICATIVE else "additive", ) model.fit(train[["ds", "y"]]) future = model.make_future_dataframe(periods=forecast_days, freq="D") fcst = model.predict(future) lbl = f"yhat_{y}y" sub = fcst[["ds", "yhat", "yhat_upper", "yhat_lower"]].rename( columns={"yhat": lbl, "yhat_upper": f"{lbl}_upper", "yhat_lower": f"{lbl}_lower"}) mape = calculate_mape(np.exp(train["y"]), np.exp(model.predict(train[["ds"]])["yhat"])) forecasts.append(dict(forecast=sub, label=lbl, mape=mape)) if not forecasts: return None eps = 1e-6 total_inv = sum(1/(f["mape"]+eps) for f in forecasts) for f in forecasts: f["weight"] = (1/(f["mape"]+eps)) / total_inv weighted = [] for f in forecasts: df_f = f["forecast"].copy() w = f["weight"] df_f.iloc[:, 1:] *= w weighted.append(df_f) ens = reduce(lambda l, r: pd.merge(l, r, on="ds", how="outer"), weighted) ens = ens.sort_values("ds").ffill().bfill() y_cols = [c for c in ens.columns if re.match(r"^yhat_.*[^_](?<!upper|lower)$", c)] up_cols = [c for c in ens.columns if "_upper" in c] lo_cols = [c for c in ens.columns if "_lower" in c] ens["yhat_log"] = ens[y_cols].sum(axis=1) ens["yhat_upper"] = np.exp(ens[up_cols].sum(axis=1)) ens["yhat_lower"] = np.exp(ens[lo_cols].sum(axis=1)) ens["yhat"] = np.exp(ens["yhat_log"]) plot_forecast(ens, df, ticker) last = ens[ens["ds"] > pd.to_datetime(end_date)].iloc[-1] cur = float(df["y_orig"].iloc[-1]) pred = float(last["yhat"]) upper, lower = float(last["yhat_upper"]), float(last["yhat_lower"]) trend = (pred - cur)/cur*100 up_ret = (upper - cur)/cur*100 lo_ret = (lower - cur)/cur*100 mar = up_ret / abs(lo_ret) if lo_ret else np.nan avg_mape = np.average([f["mape"] for f in forecasts], weights=[f["weight"] for f in forecasts]) return dict( Ticker=ticker, **{"Short Name": short_name}, **{"Current Price": cur, "Pred. Price": pred, "Trend Change%": trend, "Trend Direction": get_trend_direction(pred, cur), "Price Upper": upper, "Price Lower": lower, "Return Upper%": up_ret, "Return Lower%": lo_ret, "Predicted MAR": mar, "Avg MAPE%": avg_mape, "Accuracy": get_mape_assessment(avg_mape)} ) # -------- ランキング (変更なし) ---------- def apply_balance_score_ranking(df): # TO DO: implement this function-
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.