import gradio as gr
import pickle
from transformers import pipeline

def load_model(selected_model):
    with open(selected_model, 'rb') as file:
        loaded_model = pickle.load(file)
    return loaded_model

encoder = {
        'negative':'assets/negative.jpeg',
        'neutral':'assets/neutral.jpeg',
        'positive':'assets/positive.jpeg'
    }

def predict(model, text):
    
    selected_model = None
    with open('vectorizer.pkl', 'rb') as file:
        vectorizer = pickle.load(file)

    if 'Random Forest' == model:
        selected_model = "models/rf_twitter.pkl"
    elif 'Logistic Regression' == model:
        selected_model = "models/lg_twitter.pkl"
    elif 'Naive Bayes' == model:
        selected_model = "models/nb_twitter.pkl"
    elif 'Decision Tree' == model:
        selected_model = "models/dt_twitter.pkl"
    elif 'KNN' == model:
        selected_model = "models/knn_twitter.pkl"
    else:
        selected_model = "models/lg_twitter.pkl"
    loaded_model = load_model(selected_model)
    text_vector = vectorizer.transform([text])
    prediction = loaded_model.predict(text_vector)
    return encoder[prediction[0]]
    
classifier = pipeline(task="zero-shot-classification", model="facebook/bart-large-mnli")
def analyze_sentiment(text):
  results = classifier(text,["positive","negative",'neutral'],multi_label=True)
  mx = max(results['scores'])
  ind = results['scores'].index(mx)
  result = results['labels'][ind]
  return encoder[result]

# models = gr.Radio(['Random Forest', 'Logistic Regression','Naive Bayes','Decision Tree','KNN'], label="Choose model")
# demo = gr.Interface(fn=predict, inputs=[models,"text"], outputs="image", title="Sentiment Analysis") 
demo = gr.Interface(fn=analyze_sentiment, inputs="text", outputs="image", title="Sentiment Analysis") 
demo.launch(share=True)