# import sklearn
from os import O_ACCMODE
import gradio as gr
import joblib
from transformers import pipeline
import requests.exceptions
from huggingface_hub import HfApi, hf_hub_download
from huggingface_hub.repocard import metadata_load


app = gr.Blocks()

model_id_1 = "nlptown/bert-base-multilingual-uncased-sentiment"
model_id_2 = "juliensimon/distilbert-amazon-shoe-reviews"

def load_agent(model_id):
    """
    This function load the agent's results
    """
    # Load the metrics
    metadata = get_metadata(model_id)

    # get predictions
    predictions = predict(model_id)


    return model_id, predictions


def get_metadata(model_id):
    """
    Get the metadata of the model repo
    :param model_id:
    :return: metadata
    """
    try:
        readme_path = hf_hub_download(model_id, filename="README.md")
        metadata = metadata_load(readme_path)
        print(metadata)
        return metadata
    except requests.exceptions.HTTPError:
        return None

def get_prediction(model_id):

    classifier = pipeline("text-classification", model=model_id)
             
    def predict(review):
            prediction = classifier(review)
            print(prediction)
            stars = prediction[0]['label']
            stars = (int)(stars.split('_')[1])+1
            score = 100*prediction[0]['score']
            return "{} {:.0f}%".format("\U00002B50"*stars, score)
    return predict

with app:
    gr.Markdown(
    """
    # Compare Sentiment Analysis Models 
    
    Type text to predict sentiment.
    """)   
    with gr.Row():
        inp_1= gr.Textbox(label="Type text here.",placeholder="The customer service was satisfactory.")

    gr.Markdown(
    """
    **Model Predictions**
    """)

    gr.Markdown(
    """
    Model 1 = nlptown/bert-base-multilingual-uncased-sentiment
    """)

    with gr.Row():
        btn1 = gr.Button("Predict for Model 1")
    with gr.Row():
        out_1 = gr.Textbox(label="Prediction for Model 1")

    # classifier = pipeline("text-classification", model)   
    btn1.click(fn=get_prediction(model_id_1), inputs=inp_1, outputs=out_1)

    gr.Markdown(
    """
    Model 2 = juliensimon/distilbert-amazon-shoe-reviews
    """)

    with gr.Row():
        btn2 = gr.Button("Predict for Model 2")
    with gr.Row():
        out_2 = gr.Textbox(label="Prediction for Model 2")   
    classifier = pipeline("text-classification", model=model_id_2)    
    btn2.click(fn=get_prediction(model_id_2), inputs=inp_1, outputs=out_2)
    
app.launch()