app.py

import streamlit as st
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
import torch 
import pandas as pd 

# Function to load the pre-trained model
def load_model(model_name):
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    sentiment_pipeline = pipeline("sentiment-analysis", tokenizer=tokenizer, model=model)
    return sentiment_pipeline

# Function to load the pre-trained model
def load_finetune_model(model_name):
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    return tokenizer, model

# 
def score(item):
    return item['score']

# Streamlit app
st.title("Basic Sentiment Analysis App based on DistilBERT -- from hugging-face spaces ")
st.write("Enter a text and select a pre-trained model to get the sentiment analysis.")

# Input text
default_text = "I love my dog, she's so cute."
text = st.text_input("Enter your text:", value=default_text)

# Model selection 
# distilbert loaded from hugging face and finetuned model built on training data 
model_option = {
    "distilbert-base-uncased-finetuned-sst-2-english": {
        "labels": ["NEGATIVE", "POSITIVE"],
        "description": "This model classifies text into positive or negative sentiment. It is based on DistilBERT and fine-tuned on the Stanford Sentiment Treebank (SST-2) dataset.",
    },
    "emmaenglish/finetuned_distilbert": {
        "description": "This model detects different types of toxicity like threats, obscenity, insults, and identity-based hate in text.",
    },
}
# user choses model 
model = st.selectbox("Choose a fine-tuned model:", model_option)
# app displays model information description 
st.write("### Model Information")
st.write(f"**Description:** {model_option[model]['description']}")

# Load the model and perform sentiment analysis
if st.button("Analyze"):
    # no text is entered 
    if not text:
        st.write("Please enter a text.")
    else:
        with st.spinner("Analyzing toxicity..."):
            # user choses finetuned model trained on data in google cola b
            if model == "emmaenglish/finetuned_distilbert":
                classifier = AutoModelForSequenceClassification.from_pretrained(model)
                # tokenizer seperates text into smaller units 
                tokenizer = AutoTokenizer.from_pretrained(model)
                text_token = tokenizer(text, return_tensors="pt")
                output = classifier(**text_token)
                prediction = torch.sigmoid(output.logits)*100
                prediction = prediction.detach().numpy().tolist()[0]
                category_names = ["toxic", "severe toxic", "obscene", "threat", "insult", "identity hate"]
                output = []
                for predict, category_names in (zip(prediction, category_names)):
                    output.append({'label': category_names, 'score': predict})
                labels = output
                labels.sort(key=score, reverse=True)
                # adding catagorical data for more indepth analysis 
                df = pd.DataFrame([(text, labels[0]['label'], f"{round(labels[0]['score'], 3)}%", labels[1]['label'], f"{round(labels[1]['score'], 3)}%")], columns=('tweet/text','label 1', 'score 1', 'label 2', 'score 2'))
                st.table(df)
        
            
            else:
                # user chooses sentiment analysis of the model, no extranous model implementation nessasary 
                classifier = pipeline(model=model)
                sentiment = classifier(text)[0]["label"]
                score = classifier(text)[0]['score']
                st.write(f"The sentiment is {sentiment}.")
                st.write(f"The accuracty of this sentiment is {score}.")

else:
    # nothing has been written yet auto display 
    st.write("Enter a text and click 'Analyze' to perform toxicity analysis.")