app.py

from transformers import pipeline
import streamlit as st
import streamlit.components.v1 as components

# Load the models
pipe_1 = pipeline("text-classification", model="mavinsao/roberta-base-finetuned-mental-health")
pipe_2 = pipeline("text-classification", model="mavinsao/mi-roberta-base-finetuned-mental-health")

# Function for ensemble prediction
def ensemble_predict(text):
    # Store results from each model
    results_1 = pipe_1(text)    
    results_2 = pipe_2(text)

    # Initialize a dictionary with all potential labels to ensure they are considered
    ensemble_scores = {}

    # Add all labels from the first model's output
    for result in results_1:
        ensemble_scores[result['label']] = 0

    # Add all labels from the second model's output
    for result in results_2:
        ensemble_scores[result['label']] = 0

    # Aggregate scores from both models
    for results in [results_1, results_2]:  
        for result in results:
            label = result['label']
            score = result['score']
            ensemble_scores[label] += score / 2  # Averaging the scores

    # Determine the predicted label and confidence
    predicted_label = max(ensemble_scores, key=ensemble_scores.get) 
    confidence = ensemble_scores[predicted_label]  # Ensemble confidence

    return predicted_label, confidence

# Streamlit app
st.title('Mental Illness Prediction')

# Input text area for user input
sentence = st.text_area("Enter the long sentence to predict your mental illness state:")

if st.button('Predict'):
    # Perform the prediction
    predicted_label, confidence = ensemble_predict(sentence)

    # CSS injection to target the labels
    st.markdown("""
    <style>
        div[data-testid="metric-container"]  {
            font-weight: bold;
            font-size: 18px; /* Adjust the font size as desired */
        }
    </style>
    """, unsafe_allow_html=True)  

    # Display the result
    st.write("Result:", predicted_label)
    st.write("Confidence:", confidence)