import numpy as np import pandas as pd import streamlit as st import joblib from pathlib import Path st.set_page_config( page_title='Bike Sharing Demand (XGBoost)', page_icon='🚲', layout='centered' ) BASE_DIR = Path(__file__).resolve().parent MODEL_PATH = BASE_DIR / 'xgb_model.pkl' FEATURE_COLS_BASE = [ 'datetime', 'season', 'holiday', 'workingday', 'weather', 'temp', 'atemp', 'humidity', 'windspeed' ] @st.cache_resource def load_artifact(): if not MODEL_PATH.exists(): raise FileNotFoundError( f"Model not found: {MODEL_PATH.name}. Put '{MODEL_PATH.name}' in the repo root (same folder as app.py)." ) obj = joblib.load(MODEL_PATH) # If saved as dict: {"model": ..., "feature_cols": [...]} if isinstance(obj, dict) and 'model' in obj: model = obj['model'] feature_cols = obj.get('feature_cols', None) return model, feature_cols, True # Otherwise: directly the model return obj, None, False def add_features(df: pd.DataFrame) -> pd.DataFrame: df = df.copy() df['datetime'] = pd.to_datetime(df['datetime'], errors='coerce') if df['datetime'].isna().any(): bad = df[df['datetime'].isna()].index.tolist()[:5] raise ValueError( f"Some 'datetime' values could not be parsed. Bad rows example index: {bad}. " f"Use format like 2011-01-20 00:00:00" ) df['hour'] = df['datetime'].dt.hour.astype(int) df['weekday'] = df['datetime'].dt.weekday.astype(int) df['month'] = df['datetime'].dt.month.astype(int) df['year'] = df['datetime'].dt.year.astype(int) # engineered features (same as your notebook) df['frequently'] = df['hour'].isin([6, 7, 8, 9, 10, 13, 14, 15, 16, 17, 18]).astype(int) df['weekend'] = (df['weekday'] >= 5).astype(int) df['windy'] = (df['windspeed'] > 25).astype(int) df['best_temp'] = ((df['temp'] >= 15) & (df['temp'] <= 30)).astype(int) return df def prepare_x(df_feat: pd.DataFrame, feature_cols: list | None) -> pd.DataFrame: # drop datetime (your training X did not include datetime) x = df_feat.drop(columns=['datetime']) if feature_cols is None: return x missing = [c for c in feature_cols if c not in x.columns] extra = [c for c in x.columns if c not in feature_cols] if missing: raise ValueError(f"Missing features for model: {missing}") # drop extras if any if extra: x = x.drop(columns=extra) # reorder exactly like training x = x[feature_cols] return x def predict_counts(model, x: pd.DataFrame) -> np.ndarray: pred = model.predict(x) pred = np.clip(pred, 0, None) pred = np.rint(pred).astype(int) return pred # --- UI --- st.title('🚲 Bike Sharing Demand Forecast') st.write('Predict hourly bike rental demand using an XGBoost regression model.') model, feature_cols, is_artifact = load_artifact() with st.expander('Model info', expanded=False): st.write(f"Model file: `{MODEL_PATH.name}`") st.write(f"Loaded as artifact dict: `{is_artifact}`") if feature_cols is not None: st.write(f"Number of model features: `{len(feature_cols)}`") st.code(', '.join(feature_cols)) tab1, tab2 = st.tabs(['Single Prediction', 'Batch Prediction (CSV)']) # --- Single prediction --- with tab1: st.subheader('Single Prediction') show_debug = st.checkbox('Show debug info', value=False) dt = st.text_input('datetime (e.g. 2011-01-20 08:00:00)', value='2011-01-20 08:00:00') c1, c2 = st.columns(2) with c1: season = st.selectbox('season (1=spring, 2=summer, 3=fall, 4=winter)', [1, 2, 3, 4], index=0) holiday = st.selectbox('holiday', [0, 1], index=0) workingday = st.selectbox('workingday', [0, 1], index=1) weather = st.selectbox('weather (1-4)', [1, 2, 3, 4], index=0) with c2: temp = st.number_input('temp (°C)', value=20.0) atemp = st.number_input('atemp (°C)', value=22.0) humidity = st.number_input('humidity (0-100)', value=60.0, min_value=0.0, max_value=100.0) windspeed = st.number_input('windspeed', value=10.0, min_value=0.0) if st.button('Predict'): try: row = pd.DataFrame([{ 'datetime': dt, 'season': int(season), 'holiday': int(holiday), 'workingday': int(workingday), 'weather': int(weather), 'temp': float(temp), 'atemp': float(atemp), 'humidity': float(humidity), 'windspeed': float(windspeed), }]) feat = add_features(row) x_row = prepare_x(feat, feature_cols) pred_int = predict_counts(model, x_row)[0] st.success(f"Predicted rentals (count): **{pred_int}**") if show_debug: debug_df = feat.copy() debug_df['prediction'] = pred_int st.dataframe(debug_df) except Exception as e: st.error(str(e)) # --- Batch prediction --- with tab2: st.subheader('Batch Prediction (CSV Upload)') st.write("Upload a CSV like Kaggle `test.csv` with these columns:") st.code(', '.join(FEATURE_COLS_BASE)) file = st.file_uploader('Upload CSV', type=['csv']) if file is not None: try: df_in = pd.read_csv(file) missing_base = [c for c in FEATURE_COLS_BASE if c not in df_in.columns] if missing_base: st.error(f"Missing columns: {missing_base}") else: df_in = df_in[FEATURE_COLS_BASE].copy() df_feat = add_features(df_in) x_batch = prepare_x(df_feat, feature_cols) preds = predict_counts(model, x_batch) out = pd.DataFrame({ 'datetime': df_feat['datetime'], 'count': preds }) st.success('Predictions created!') st.dataframe(out.head(20)) csv_bytes = out.to_csv(index=False).encode('utf-8') st.download_button( label='Download predictions.csv', data=csv_bytes, file_name='predictions.csv', mime='text/csv' ) except Exception as e: st.error(str(e)) st.caption("Model file expected: xgb_model.pkl (repo root).")