import gradio as gr
import hopsworks
from datasets import load_dataset
import pandas as pd

project = hopsworks.login()
fs = project.get_feature_store()

dataset_api = project.get_dataset_api()

dataset = load_dataset("torileatherman/sentiment_analysis_batch_predictions", split='train')
predictions_df = pd.DataFrame(dataset)
predictions_df_url0 = predictions_df['Url'].iloc[0]
predictions_df_url1 = predictions_df['Url'].iloc[1]
predictions_df_url2 = predictions_df['Url'].iloc[2]
predictions_df_urls = [[predictions_df_url0],
                       [predictions_df_url1],
                       [predictions_df_url2]]

def article_selection(sentiment):
    if sentiment == "Positive":
            return predictions_df_urls #f"""The sentence you requested is Positive!"""
    elif sentiment == "Negative":
        return f"""The sentence you requested is Negative!""" 
    else:
        return f"""The sentence you requested is Neutral!""" 

description =  '''
        This application recommends news articles depending on the sentiment of the headline.
        Enter your preference of what type of news articles you would like recommended to you today: Positive, Negative, or Neutral.
        '''

demo = gr.Interface(
    fn=article_selection,
    title = 'Extractive News Summarizer BART',
    inputs = gr.Dropdown(["Positive","Negative","Neutral"], label="What type of news articles would you like recommended?"),
    outputs = gr.Textbox(label="Recommended News Articles", lines=3),
    description = description,
    examples=predictions_df_urls)


#TODO 
#demo = gr.TabbedInterface([url_demo, voice_demo], ["Swedish YouTube Video to English Text", "Swedish Audio to English Text"])


demo.launch()