app.py

import gradio as gr
import pandas as pd
import pickle
from sklearn.metrics.pairwise import cosine_similarity
import heapq

# Load data and model
df = pd.read_csv('./DATA/spotify_millsongdata.csv')

# Load saved embeddings
with open("./DATA/lyrics_embeddings.pkl", "rb") as f:
    lyrics_embeddings = pickle.load(f)

# List of artists and songs
artists = df['artist'].unique()
song_titles = df['song']

# Recommendation logic
def recommend_songs(song_index, top_n=5, batch_size=100):
    top_sim_scores = []
    num_batches = len(df) // batch_size + 1

    for i in range(num_batches):
        start_idx = i * batch_size
        end_idx = min((i + 1) * batch_size, len(df))

        # Compute cosine similarity for the current batch
        cosine_sim_batch = cosine_similarity(
            lyrics_embeddings[start_idx:end_idx],
            [lyrics_embeddings[song_index]]
        )

        # Select the top N most similar songs
        for j, sim_score in enumerate(cosine_sim_batch):
            global_idx = start_idx + j
            heapq.heappush(top_sim_scores, (sim_score[0], global_idx))
            if len(top_sim_scores) > top_n + 1:
                heapq.heappop(top_sim_scores)

    # Exclude the selected song itself and return the most similar songs with their similarity scores
    top_sim_scores = sorted(top_sim_scores, key=lambda x: x[0], reverse=True)[1:top_n+1]
    recommended_songs = [(song_titles[i[1]], df['link'][i[1]], round(i[0], 2)) for i in top_sim_scores]

    return recommended_songs

# Interface logic function
def get_songs_by_artist(artist_name):
    filtered_songs = df[df['artist'] == artist_name]['song'].tolist()
    return gr.update(choices=filtered_songs, value=filtered_songs[0] if filtered_songs else None)


def gradio_recommend(song_title):
    try:
        # Find the index of the selected song
        song_index = song_titles[song_titles == song_title].index[0]
        # Get recommended songs
        recommendations = recommend_songs(song_index)

        # Format the output, making song links clickable
        result = "<div style='text-align: left;'>"
        for song, link, sim_score in recommendations:
            result += f"<b>Song Name:</b> {song}<br>"
            result += f"<b>Search Link:</b> <a href='https://www.google.com/search?q={link}' target='_blank'>{link}</a><br>"
            result += f"<b>Lyrics Similarity:</b> {sim_score:.2f}<br><br>"
        result += "</div>"
        return result
    except IndexError:
        return "Song not found."

# Create Gradio multi-page interface
with gr.Blocks(css="""
@media (max-width: 768px) {
    .gr-container { 
        width: 100%; 
        padding: 10px; 
        box-sizing: border-box; 
    }
    .gr-dropdown select {
    width: 100%;
    height: 40px; /* Limit height */
    font-size: 16px;
    padding: 5px;
    box-sizing: border-box;
    }
    .gr-button { 
        width: 100%; 
        font-size: 16px; 
        margin-top: 10px; 
    }
    .gr-html, .gr-row { 
        width: 100%; 
        font-size: 16px; 
        margin: 10px 0; 
    }
    h1 { 
        font-size: 24px; 
    }
    p { 
        font-size: 14px; 
    }
    .gr-dropdown::after {
    content: '';
    width: 12px;
    height: 12px;
    border: solid black;
    border-width: 0 2px 2px 0;
    display: inline-block;
    transform: rotate(45deg);
    margin-left: 10px;
   }
}
""") as demo:

    gr.Markdown(
        """
        <div style="text-align: center; padding: 20px;">
            <h1 style="color: #1DB954;">Music Recommendation System</h1>
            <p style="font-size: 18px;">Get the most relevant song recommendations based on lyrics similarity</p>
        </div>
        """
    )

    # Page 1: Select artist
    with gr.Row():
        with gr.Column():
          artist_dropdown = gr.Dropdown(choices=list(artists), label="Select Artist")
          next_button = gr.Button("Next")

    # Page 2: Select song and get recommendations
    with gr.Row(visible=False) as song_selection_row:
        song_dropdown = gr.Dropdown(label="Select Song")
        recommend_button = gr.Button("Get Recommendations")
        output = gr.HTML(label="Recommended Similar Songs")

    # Event bindings
    artist_dropdown.change(get_songs_by_artist, inputs=artist_dropdown, outputs=song_dropdown)
    next_button.click(lambda: gr.update(visible=True), None, song_selection_row)
    recommend_button.click(gradio_recommend, inputs=song_dropdown, outputs=output)

if __name__ == "__main__":
    demo.launch()