import streamlit as st  #Web App
from transformers import pipeline
from transformers import AutoTokenizer, AutoModelForSequenceClassification


#title
st.title("Sentiment Analysis")


def analyze(input, model):
    return "This is a sample output"


# load my fine-tuned model
fine_tuned = "jbraha/tweet-bert"
labels = {'LABEL_0': 'toxic', 'LABEL_1': 'severe_toxic', 'LABEL_2': 'obscene', 'LABEL_3': 'threat',
          'LABEL_4': 'insult', 'LABEL_5': 'identity_hate'}

# make a dictionary of the labels with keys like "LABEL_0" and values like "toxic"

#text insert
input = st.text_area("Insert text to be analyzed", value="Nice to see you today.", 
                     height=None, max_chars=None, key=None, help=None, on_change=None, 
                     args=None, kwargs=None, placeholder=None, disabled=False, 
                     label_visibility="visible")

option = st.selectbox(
    'Choose a transformer model:',
    ('Default', 'Fine-Tuned' , 'Roberta'))


if option == 'Fine-Tuned':
    model = AutoModelForSequenceClassification.from_pretrained(fine_tuned)
    tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
    classifier = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
elif option == 'Roberta':
    model = AutoModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
    tokenizer = AutoTokenizer.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
    classifier = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer, top_k=None)
else:
    classifier = pipeline('sentiment-analysis')


if st.button('Analyze'):
    result = classifier(input)
    print(result)
    print(type(result))
    output = None
    result = result[0]
    if option == 'Fine-Tuned':
        output = {'Toxic': result['LABEL_0']}
        del result['LABEL_0']
        output[max(result, key=result.get)] = result[max(result, key=result.get)]
    else:
        output = result
    st.write(output)
else:
    st.write('Excited to analyze!')