import streamlit as st
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing.sequence import pad_sequences
import joblib
import pandas as pd
import numpy as np

# Load your models
emotion_model = load_model('lstm_model.h5')
recommender_model = np.load('knn_model.npy', allow_pickle=True)

# Load the tokenizer (if used during training)
# tokenizer = joblib.load('tokenizer.pkl')  # Update with actual file name

# Load the dataset
df = pd.read_csv('df1.csv')  # Make sure this is the correct DataFrame

# Set up the title of the app
st.title('Emotion and Audio Feature-based Song Recommendation System')

# Input field for lyrics
st.header('Enter Song Lyrics')
lyrics = st.text_area("Input the lyrics of the song here:")

# Input fields for audio features
st.header('Enter Audio Features')
audio_features = []
for feature_name in df.columns:  # Make sure this matches your DataFrame
    feature = st.number_input(f"Enter value for {feature_name}:", step=0.01)
    audio_features.append(feature)

# Predict and Recommend button
if st.button('Predict Emotion and Recommend Songs'):
    if lyrics and all(audio_features):
        sequence = tokenizer.texts_to_sequences([lyrics])
        padded_sequence = pad_sequences(sequence, maxlen=128)
        emotion = emotion_model.predict(padded_sequence).flatten()  # Flatten if needed

        # Combine emotion and audio features for recommendation
        combined_features = np.concatenate([[emotion], audio_features])

        # Generate recommendations using the KNN model
        distances, indices = recommender_model.kneighbors([combined_features], n_neighbors=5)
        recommended_songs = df.iloc[indices.flatten()]

        # Display emotion and recommendations
        st.write("Emotion Detected:", emotion[0])  # Adjust as per your model's output
        st.header('Recommended Songs')
        for _, song in recommended_songs.iterrows():
            st.write(song)  # Adjust based on your dataset
    else:
        st.error("Please fill in all the fields.")