app.py

from mabwiser.mab import MAB, LearningPolicy

def bandit_factory(bandit_type, arms):
  if bandit_type == "Epsilon Greedy":
    result = MAB(arms=arms,
             learning_policy=LearningPolicy.EpsilonGreedy(epsilon=0.3),
             seed=1234)
  elif bandit_type == "UCB":
    result = MAB(arms=arms,
             learning_policy=LearningPolicy.UCB1(alpha=1),
             seed=1234)
  elif bandit_type == "Non-Stationary":
    result = NSBandit(arms=arms, epsilon=0.3, alpha=0.2)
  else:
    raise ValueError("Invalid bandit type")

  result.partial_fit(decisions=arms, rewards=[3]*len(arms))
  return result

class NSBandit:
  def __init__(self, arms, epsilon, alpha):
    self.arms = arms
    self.epsilon = epsilon
    self.alpha = alpha
    self.means = {arm: None for arm in arms}
    self.t = 0

  def partial_fit(self, decisions, rewards):
    for arm, reward in zip(decisions, rewards):
      if self.means[arm] is None:
        self.means[arm] = reward
      else:
        self.means[arm] += self.alpha * (reward - self.means[arm])
    self.t += 1


  def predict(self):
    nones = [t[0] for t in self.means.items() if t[1] is None]
    if len(nones) > 0:
      return random.choice(nones)

    best = max(self.means, key=self.means.get)

    if random.random() < self.epsilon:
      return random.choice(list(set(self.arms) - {best}))
    else:
      return max(self.means, key=self.means.get)

import gradio as gr
import random
import pandas as pd

# Load songs dataset
#file_path = "/content/drive/My Drive/MIT/RealTime/songs_single_genre.csv"
#file_path = "/content/songs_single_genre.csv"
file_path = "songs_final.csv"
song_df = pd.read_csv(file_path)
song_df.rename(columns={"name": "song"}, inplace=True)

# Drop missing values in key columns
song_df = song_df.dropna(subset=['genre', 'song', 'artist', 'img', 'preview'])
song_df = song_df[song_df['genre'].isin(['Hip-Hop', 'Pop', 'Reggaeton','Country'])]

# Get unique genres
genres = list(song_df['genre'].unique())

# Create a dictionary to store song details, including the image and preview
songs = {}
for genre in genres:
    songs[genre] = list(song_df[song_df['genre'] == genre][['song', 'artist', 'img', 'preview']].itertuples(index=False))

# Initialize bandit
bandit = bandit_factory("Epsilon Greedy", list(songs.keys()))

# Initial song, genre, and media
initial_genre = "Pop"
initial_song, initial_artist, initial_img, initial_preview = random.choice(songs[initial_genre])
initial_song_display = f"{initial_song}\nBy {initial_artist}"

# Function to handle rating submission
def submit_rating(cur_genre, rating, MAB):
    # Adjust MAB
    MAB.partial_fit(decisions=[cur_genre], rewards=[rating])

    # Get new genre
    new_genre = MAB.predict()

    # Select a random song from the new genre
    new_song, new_artist, new_img, new_preview = random.choice(songs[new_genre])

    # Format song text
    new_song_display = f"{new_song}\nBy {new_artist}"

    return new_genre, new_song_display, new_img, new_preview, MAB

# Function to change bandit model
def change_bandit(bandit_type):
    return bandit_factory(bandit_type, list(songs.keys()))

# Gradio UI
with gr.Blocks() as demo:
    # State variables
    MAB = gr.State(bandit)
    cur_genre = gr.State(initial_genre)

    model = gr.Dropdown(
        ["Epsilon Greedy", "UCB", "Non-Stationary"], label="Bandit Model", interactive=True
    )

    # Image output component for album cover
    image_output = gr.Image(value=initial_img, label="Album Cover", interactive=False)

    # Text output component for song details
    text_output = gr.Textbox(value=initial_song_display, label="Song", interactive=False)

    # Audio preview component
    audio_output = gr.Audio(value=initial_preview, label="Preview", interactive=False)

    # Slider for rating
    slider = gr.Slider(minimum=1, maximum=5, step=1, label="Rate the song (1-5)")

    # Submit button
    submit_btn = gr.Button("Submit")

    # Change bandit model
    model.input(
        fn=change_bandit,
        inputs=model,
        outputs=MAB
    )

    # Submit button action
    submit_btn.click(
        fn=submit_rating,
        inputs=[cur_genre, slider, MAB],
        outputs=[cur_genre, text_output, image_output, audio_output, MAB]
    )

demo.launch()