import torch
from transformers import AutoModel
import torch.nn as nn
from PIL import Image
import numpy as np
import streamlit as st

# Set the device
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Load the trained model from the Hugging Face Hub
model = AutoModel.from_pretrained('dhhd255/parkinsons_pred0.1')

# Move the model to the device
model = model.to(device)

# Add custom CSS to use the Inter font, define custom classes for healthy and parkinsons results, increase the font size, make the text bold, and define the footer styles
st.markdown("""
 <style>
 @import url('https://fonts.googleapis.com/css2?family=Inter&display=swap');
 body {
 font-family: 'Inter', sans-serif;
 }
 .result {
 font-size: 24px;
 font-weight: bold;
 }
 .healthy {
 color: #007E3F;
 }
 .parkinsons {
 color: #C30000;
 }
 .caption_c{
  position: relative;
  display: flex;
  flex-directon: column;
  align-items: center;
  top: calc(99vh - 370px);
 }
 .caption {
  
 text-align: center;
 color: #646464;
 font-size: 14px;
 }
  button:hover {
 background-color: lightblue !important;
 outline-color: lightblue !important;
 }
 button:focus {
 background-color: lightblue !important;
 outline-color: lightblue !important;
 }
 </style>
""", unsafe_allow_html=True)

st.title("Parkinson's Disease Prediction")

uploaded_file = st.file_uploader("Upload your :blue[Spiral] drawing here", type=["png", "jpg", "jpeg"])
if uploaded_file is not None:
    col1, col2 = st.columns(2)

    # Load and resize the image
    image_size = (224, 224)
    new_image = Image.open(uploaded_file).convert('RGB').resize(image_size)
    col1.image(new_image, use_column_width=True)
    new_image = np.array(new_image)
    new_image = torch.from_numpy(new_image).transpose(0, 2).float().unsqueeze(0)

    # Move the data to the device
    new_image = new_image.to(device)

    # Make predictions using the trained model
    with torch.no_grad():
        predictions = model(new_image)
        logits = predictions.last_hidden_state
        logits = logits.view(logits.shape[0], -1)
        num_classes=2
        feature_reducer = nn.Linear(logits.shape[1], num_classes)

        logits = logits.to(device)
        feature_reducer = feature_reducer.to(device)

        logits = feature_reducer(logits)
        predicted_class = torch.argmax(logits, dim=1).item()
        confidence = torch.softmax(logits, dim=1)[0][predicted_class].item()
        if(predicted_class == 0):
            col2.markdown('<span class="result parkinsons">Predicted class: Parkinson\'s</span>', unsafe_allow_html=True)
            col2.caption(f'{confidence*100:.0f}% sure')
        else:
            col2.markdown('<span class="result healthy">Predicted class: Healthy</span>', unsafe_allow_html=True)
            col2.caption(f'{confidence*100:.0f}% sure')

# Add a caption at the bottom of the page
st.markdown('<div class="caption_c"><p class="caption">Made with love by Jayant</p></div>', unsafe_allow_html=True)