import numpy as np import pandas as pd import streamlit as st import joblib from pathlib import Path # -------------------------------------------------- # Page config # -------------------------------------------------- st.set_page_config( page_title='BPM Predictor (LightGBM)', page_icon='🎵', layout='centered' ) st.title('🎵 BPM Predictor') st.write('Predict **Beats Per Minute (BPM)** from audio features using a trained **LightGBM** regression model.') BASE_DIR = Path(__file__).resolve().parent MODEL_PATH = BASE_DIR / 'lgbm_model.pkl' FEATURES_PATH = BASE_DIR / 'feature_names.pkl' # -------------------------------------------------- # Load artifacts # -------------------------------------------------- @st.cache_resource def load_artifacts(): if not MODEL_PATH.exists(): raise FileNotFoundError( f'Model not found: {MODEL_PATH.name}. Put it in the repo root (same folder as app.py).' ) if not FEATURES_PATH.exists(): raise FileNotFoundError( f'Feature list not found: {FEATURES_PATH.name}. Put it in the repo root (same folder as app.py).' ) model = joblib.load(MODEL_PATH) feature_names = joblib.load(FEATURES_PATH) if not isinstance(feature_names, list) or len(feature_names) == 0: raise ValueError('feature_names.pkl must contain a non-empty list of feature names.') return model, feature_names model, feature_names = load_artifacts() # -------------------------------------------------- # Feature engineering (MUST match training) # -------------------------------------------------- def add_features(df: pd.DataFrame) -> pd.DataFrame: df = df.copy() # required base columns required = [ 'RhythmScore', 'AudioLoudness', 'VocalContent', 'AcousticQuality', 'InstrumentalScore', 'LivePerformanceLikelihood', 'MoodScore', 'TrackDurationMs', 'Energy' ] missing = [c for c in required if c not in df.columns] if missing: raise ValueError(f'Missing required base columns: {missing}') # engineered columns (same as training) df['Rhythm_Energy'] = df['RhythmScore'] * df['Energy'] df['Mood_Energy'] = df['MoodScore'] * df['Energy'] df['Loudness_Energy'] = df['AudioLoudness'] * df['Energy'] return df def make_single_row_input(values: dict) -> pd.DataFrame: df = pd.DataFrame([values]) df = add_features(df) # align to training columns exactly missing = [c for c in feature_names if c not in df.columns] if missing: raise ValueError(f'Missing engineered columns after feature engineering: {missing}') extra = [c for c in df.columns if c not in feature_names] if extra: df = df.drop(columns=extra) df = df[feature_names] return df # -------------------------------------------------- # UI # -------------------------------------------------- with st.expander('ℹ️ Model info', expanded=False): st.write('**Model:** LightGBM Regressor') st.write('**Target:** BeatsPerMinute') st.write('**Features:** numeric audio descriptors + interaction features') st.caption('Note: In this dataset, features show limited predictive signal, so predictions may be close to the mean BPM.') st.subheader('Enter audio feature values') col1, col2 = st.columns(2) with col1: rhythm = st.slider('RhythmScore', min_value=0.0, max_value=1.0, value=0.6, step=0.001) energy = st.slider('Energy', min_value=0.0, max_value=1.0, value=0.6, step=0.001) mood = st.slider('MoodScore', min_value=0.0, max_value=1.0, value=0.6, step=0.001) live = st.slider('LivePerformanceLikelihood', min_value=0.0, max_value=1.0, value=0.2, step=0.001) with col2: loudness = st.number_input('AudioLoudness (dB)', value=-10.0, step=0.1) vocal = st.slider('VocalContent', min_value=0.0, max_value=1.0, value=0.08, step=0.001) acoustic = st.slider('AcousticQuality', min_value=0.0, max_value=1.0, value=0.25, step=0.001) instrumental = st.slider('InstrumentalScore', min_value=0.0, max_value=1.0, value=0.10, step=0.001) track_ms = st.number_input('TrackDurationMs', value=240000, step=1000, min_value=10000) input_values = { 'RhythmScore': float(rhythm), 'AudioLoudness': float(loudness), 'VocalContent': float(vocal), 'AcousticQuality': float(acoustic), 'InstrumentalScore': float(instrumental), 'LivePerformanceLikelihood': float(live), 'MoodScore': float(mood), 'TrackDurationMs': float(track_ms), 'Energy': float(energy) } st.divider() if st.button('Predict BPM'): try: X = make_single_row_input(input_values) pred = float(model.predict(X)[0]) st.success(f'Predicted BPM: **{pred:.2f}**') with st.expander('See model input (after feature engineering)'): st.dataframe(X) except Exception as e: st.error(str(e)) st.caption('Tip: Keep `lgbm_model.pkl` and `feature_names.pkl` in the repo root next to `app.py`.')