app.py

import streamlit as st
import numpy as np
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing.sequence import pad_sequences
import joblib
import pandas as pd
from sklearn.neighbors import NearestNeighbors
from sklearn.preprocessing import StandardScaler
from sklearn.metrics.pairwise import cosine_similarity

# Load the LSTM model for emotion prediction
emotion_model = load_model('lstm_model.h5')

# Load the KNN model
knn_model = joblib.load('knn_model.joblib')

# Load the tokenizer
tokenizer = joblib.load('tokenizer.pkl')

# Load the dataset
df = pd.read_csv('df1.csv')

# Load the scaler for KNN
scaler_knn = StandardScaler()

# Function for hybrid recommendation
def hybrid_recommendation(song_index):
    # Get data for the query song
    query_data = df.iloc[song_index]

    # Process the lyrics for emotion prediction using LSTM
    sequence = tokenizer.texts_to_sequences([query_data['lyrics']])
    padded_sequence = pad_sequences(sequence, maxlen=50)
    predicted_emotion = emotion_model.predict(padded_sequence).flatten()

    # Preprocess for KNN
    audio_features_knn = query_data[['danceability', 'energy', 'key', 'loudness', 'mode', 'speechiness',
                                      'acousticness', 'instrumentalness', 'liveness', 'valence', 'tempo',
                                      'duration_ms', 'time_signature']].values.reshape(1, -1)
    mood_cats = query_data[['mood_cats']]
    mood_cats_df = pd.DataFrame(mood_cats)
    audio_features_scaled_knn = scaler_knn.fit_transform(audio_features_knn)
    combined_features = pd.concat([mood_cats_df, pd.DataFrame(audio_features_scaled_knn, columns=audio_features_knn.columns)], axis=1)
    
    # Predict using the KNN model
    knn_recommendations = knn_model.kneighbors(combined_features, n_neighbors=5, return_distance=False)[0]

    # Mapping emotion predictions to encoded categories
    emotion_mapping = {0: 'happy', 1: 'sad', 2: 'calm', 3: 'anger'}
    encoded_emotion = np.argmax(predicted_emotion)
    emotion_category = emotion_mapping[encoded_emotion]

    # Compute cosine similarity for content-based recommendation
    features_for_similarity = df[['danceability', 'energy', 'key', 'loudness', 'mode', 'speechiness',
                                  'acousticness', 'instrumentalness', 'liveness', 'valence', 'tempo',
                                  'duration_ms', 'time_signature']].values

    cosine_similarities = cosine_similarity([emotion_category], features_for_similarity).flatten()

    # Combine recommendations from both models
    combined_indices = np.argsort(-np.concatenate([knn_recommendations, cosine_similarities]))
    hybrid_recs_sorted = combined_indices[:5]  # Select top 5 recommendations

    return hybrid_recs_sorted

# Set up the title of the app
st.title('Hybrid Recommender App')

# Get song index from user input
song_index_to_recommend = st.number_input('Enter song index:', min_value=0, max_value=len(df)-1, value=0)

# Get hybrid recommendations
hybrid_recs = hybrid_recommendation(song_index_to_recommend)

# Display the recommendations
st.write("Hybrid Recommendations:")
for index in hybrid_recs:
    st.write(f"Song Index: {index}, Title: {df.iloc[index]['title']}, Artist: {df.iloc[index]['artist']}")