update

app.py

@@ -25,3 +25,49 @@ contents = os.listdir(pwd)
 print("Contents of the Directory:")
 for item in contents:
     print(item)
+
+df = pd.read_csv('v2ga_w_embeddings.parquet')
+
+def get_most_similar_songs(artist, title, df):
+    def find_most_similar(embedding_column):
+        chosen_song = df[(df['artist'] == artist) & (df['title'] == title)][embedding_column].values
+        if len(chosen_song) == 0:
+            return None
+
+        chosen_song = chosen_song.reshape(1, -1)
+        similarity_matrix = cosine_similarity(df[embedding_column].values.tolist(), chosen_song)
+        most_similar_indices = np.argsort(similarity_matrix.flatten())[-5:-1][::-1] # Top 4 excluding the selected song
+        return df.iloc[most_similar_indices][['title', 'artist', 'lyrics']].to_dict(orient='records')
+
+    results = {}
+    for embedding in ['embedding_glove', 'embedding_minilm', 'embedding_roberta', 'embedding_gpt']:
+        most_similar = find_most_similar(embedding)
+        if most_similar is None:
+            return "Song not found. Please ensure the artist and title are correct."
+
+        results[embedding] = most_similar
+
+    return results
+
+def update_titles_dropdown(artist):
+    titles = sorted(df[df['artist'] == artist]['title'].unique())
+    return titles
+
+artists = sorted(df['artist'].unique())
+
+artist_dropdown = gr.inputs.Dropdown(artists, label="Artist")
+title_dropdown = gr.inputs.Dropdown([], label="Title", updatable=True)
+
+output_interface = gr.outputs.JSON(label="Similar Songs")
+
+iface = gr.Interface(
+    fn=get_most_similar_songs,
+    inputs=[artist_dropdown, title_dropdown],
+    outputs=output_interface,
+    examples=[("The Beatles", "Let It Be"), ("Eminem", "Lose Yourself")],
+    title="Semantic Song Search: Most Similar Song",
+    description="Find the 4 most similar songs to the selected song based on different embeddings (GloVe, MiniLM, RoBERTa, GPT).",
+    update=update_titles_dropdown
+)
+
+iface.launch()