app.py

import gradio as gr
import pandas as pd
import numpy as np
import pickle
from sklearn.preprocessing import StandardScaler
import sklearn
import os

print(f"Prediction environment scikit-learn version: {sklearn.__version__}")

def decode_file(file_path):
    with open(file_path, 'rb') as file:
        obj = pickle.load(file)
    return obj

# Load the model without try/except
model = decode_file('./model.pkl')

def preprocess_input(data_dict):
    """Preprocess input data to match the training format"""
    # First create DataFrame with original columns
    df = pd.DataFrame([data_dict])
    
    # Store original features before transformation
    original_df = df.copy()
    
    # Numeric features
    numeric_features = ['age', 'avg_glucose_level', 'bmi']
    
    # Scale numeric features and add prefix
    scaler = StandardScaler()
    scaled_features = scaler.fit_transform(df[numeric_features])
    for i, feat in enumerate(numeric_features):
        df[f'num__{feat}'] = scaled_features[:, i]
    
    # Create categorical features
    # Gender
    df['cat__gender_Male'] = (df['gender'] == 'Male').astype(float)
    df['cat__gender_Other'] = 0.0
    
    # Hypertension
    df['cat__hypertension_1'] = df['hypertension'].astype(float)
    
    # Heart Disease
    df['cat__heart_disease_1'] = df['heart_disease'].astype(float)
    
    # Ever Married
    df['cat__ever_married_Yes'] = (df['ever_married'] == 'Yes').astype(float)
    
    # Work Type
    df['cat__work_type_Never_worked'] = (df['work_type'] == 'Never_worked').astype(float)
    df['cat__work_type_Private'] = (df['work_type'] == 'Private').astype(float)
    df['cat__work_type_Self-employed'] = (df['work_type'] == 'Self-employed').astype(float)
    df['cat__work_type_children'] = (df['work_type'] == 'children').astype(float)
    
    # Residence Type
    df['cat__Residence_type_Urban'] = (df['Residence_type'] == 'Urban').astype(float)
    
    # Smoking Status
    df['cat__smoking_status_formerly smoked'] = (df['smoking_status'] == 'formerly smoked').astype(float)
    df['cat__smoking_status_never smoked'] = (df['smoking_status'] == 'never smoked').astype(float)
    df['cat__smoking_status_smokes'] = (df['smoking_status'] == 'smokes').astype(float)
    
    # Select only the transformed columns in the correct order
    feature_columns = [
        'num__age', 'num__avg_glucose_level', 'num__bmi',
        'cat__gender_Male', 'cat__gender_Other', 'cat__hypertension_1',
        'cat__heart_disease_1', 'cat__ever_married_Yes',
        'cat__work_type_Never_worked', 'cat__work_type_Private',
        'cat__work_type_Self-employed', 'cat__work_type_children',
        'cat__Residence_type_Urban', 'cat__smoking_status_formerly smoked',
        'cat__smoking_status_never smoked', 'cat__smoking_status_smokes'
    ]
    
    return df[feature_columns]

def predict(gender, age, hypertension, ever_married, work_type, heart_disease, 
           avg_glucose_level, bmi, smoking_status, Residence_type):
    """Make prediction using the loaded model"""
    if model is None:
        return "Error: Model not loaded"
        
    # Create input dictionary
    input_data = {
        'gender': gender,
        'age': float(age),  # Ensure numeric
        'hypertension': 1 if hypertension == 'Yes' else 0,
        'heart_disease': 1 if heart_disease == 'Yes' else 0,
        'ever_married': ever_married,
        'work_type': work_type,
        'Residence_type': Residence_type,
        'avg_glucose_level': float(avg_glucose_level),  # Ensure numeric
        'bmi': float(bmi),  # Ensure numeric
        'smoking_status': smoking_status
    }
    
    # Preprocess the input
    processed_input = preprocess_input(input_data)
    
    # Use the loaded model
    try:
        prediction = model.predict_proba(processed_input)[0][1]
        return f"The probability of stroke is {prediction:.2%}"
    except Exception as e:
        return f"Error making prediction: {str(e)}, model is not valid" # updated here

# Create the Gradio interface
iface = gr.Interface(
    fn=predict,
    inputs=[
        gr.Radio(choices=['Female', 'Male'], label="Gender"),
        gr.Slider(minimum=0, maximum=100, label="Age"),
        gr.Radio(choices=['Yes', 'No'], label="Hypertension"),
        gr.Radio(choices=['Yes', 'No'], label="Ever Married"),
        gr.Radio(choices=['Private', 'Self-employed', 'Govt_job', 'children', 'Never_worked'], label="Work Type"),
        gr.Radio(choices=['Yes', 'No'], label="Heart Disease"),
        gr.Number(label="Average Glucose Level"),
        gr.Slider(minimum=10, maximum=50, label="BMI"),
        gr.Radio(choices=['formerly smoked', 'never smoked', 'smokes', 'Unknown'], label="Smoking Status"),
        gr.Radio(choices=['Urban', 'Rural'], label="Residence Type")
    ],
    outputs='text',
    title='Stroke Probability Predictor',
    description='Predicts the probability of having a stroke based on input features.'
)

if __name__ == "__main__":
    iface.launch()