I thin that the daily data is difficult let us group it by weeks and calculate again
- User
I thin that the daily data is difficult let us group it by weeks and calculate again
Assistant
import pandas as pd import numpy as np from prophet import Prophet import optuna # ------------------------------------------------------------------ # 1. Load your data (ds = date, y = daily contract count) # ------------------------------------------------------------------ df = pd.read_csv('your_file.csv', parse_dates=['ds']) # ------------------------------------------------------------------ # 2. Keep only business days (drop Sat/Sun) # ------------------------------------------------------------------ df = df[df['ds'].dt.weekday < 5].reset_index(drop=True) # ------------------------------------------------------------------ # 3. Group by week and calculate weekly sum # ------------------------------------------------------------------ df_weekly = df.groupby(pd.Grouper(key='ds', freq='W')).sum().reset_index() df_weekly['ds'] = df_weekly['ds'].dt.date # Keep only the date part df_weekly = df_weekly.rename(columns={'y': 'y'}) # ------------------------------------------------------------------ # 4. Train / test split -> last 365 BUSINESS weeks as hold-out # ------------------------------------------------------------------ HOLDOUT_WEEKS = 365 train_df = df_weekly.iloc[:-HOLDOUT_WEEKS] test_df = df_weekly.iloc[-HOLDOUT_WEEKS:] # ------------------------------------------------------------------ # 5. SMAPE helper # ------------------------------------------------------------------ def smape(y_true, y_pred): y_true, y_pred = np.array(y_true), np.array(y_pred) return 100 * np.mean(2 * np.abs(y_pred - y_true) / (np.abs(y_true) + np.abs(y_pred) + 1e-8)) # ------------------------------------------------------------------ # 6. Optuna objective # ------------------------------------------------------------------ def objective(trial): params = { 'changepoint_prior_scale': trial.suggest_float('changepoint_prior_scale', 0.001, 0.5, log=True), 'seasonality_prior_scale': trial.suggest_float('seasonality_prior_scale', 0.01, 10, log=True), 'changepoint_range' : trial.suggest_float('changepoint_range', 0.8, 0.95), 'n_changepoints' : trial.suggest_int('n_changepoints', 15, 50), 'seasonality_mode' : trial.suggest_categorical('seasonality_mode', ['additive', 'multiplicative']), 'interval_width' : trial.suggest_float('interval_width', 0.90, 0.95) } m = Prophet( daily_seasonality=False, weekly_seasonality=True, yearly_seasonality=True, **params ) m.fit(train_df) # Forecast exactly the test period future = pd.DataFrame({'ds': test_df['ds']}) fcst = m.predict(future) # Align and compute SMAPE merged = pd.merge(test_df, fcst[['ds', 'yhat']], on='ds', how='inner') return smape(merged['y'], merged['yhat']) # ------------------------------------------------------------------ # 7. Run Bayesian optimisation # ------------------------------------------------------------------ study = optuna.create_study(direction='minimize') study.optimize(objective, n_trials=30, show_progress_bar=True) print("Best SMAPE :", round(study.best_value, 2)) print("Best params:", study.best_params) # ------------------------------------------------------------------ # 8. Final model on best parameters # ------------------------------------------------------------------ best = study.best_params m_final = Prophet( daily_seasonality=False, weekly_seasonality=True, yearly_seasonality=True, **best ) m_final.fit(train_df) future = pd.DataFrame({'ds': test_df['ds']}) fcst = m_final.predict(future) merged = pd.merge(test_df, fcst[['ds', 'yhat', 'yhat_lower', 'yhat_upper']], on='ds') final_smape = smape(merged['y'], merged['yhat']) print("Final back-test SMAPE:", round(final_smape, 2)) # ------------------------------------------------------------------ # 9. Quick plot # ------------------------------------------------------------------ import matplotlib.pyplot as plt plt.figure(figsize=(14,4)) plt.plot(merged['ds'], merged['y'], label='actual') plt.plot(merged['ds'], merged['yhat'], label='forecast') plt.fill_between(merged['ds'], merged['yhat_lower'], merged['yhat_upper'], alpha=.2) plt.axvline(train_df['ds'].iloc[-1], color='k', ls='--') plt.legend(); plt.title(f"Back-test SMAPE = {final_smape:.2f}%"); plt.show()The key change is grouping by week using
pd.Grouper(key='ds', freq='W')and then summing the values within each week. This provides a more stable time series for Prophet to learn from, especially when dealing with sparse data like weekly counts.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.