import streamlit as st
from tensorflow.keras.models import load_model
import pickle
import numpy as np
import os
import google.generativeai as genai

# Configure the generative AI with the API key
genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))

# Load the saved models
with open('rf_model.pkl', 'rb') as file:
    rf_model = pickle.load(file)

deep_model = load_model('deep_model.h5')

# Define the function to make predictions using the RandomForestRegressor model
def make_rf_prediction(model, input_data):
    prediction = model.predict(input_data)
    return prediction

# Define the function to make predictions using the deep learning model
def make_deep_prediction(model, input_data):
    prediction = model.predict(input_data).flatten()
    return prediction

# Define the function to calculate GPA
def calculate_gpa(total_score):
    if total_score >= 70:
        return 'A (5 points)', total_score
    elif total_score >= 60:
        return 'B (4 points)', total_score
    elif total_score >= 50:
        return 'C (3 points)', total_score
    elif total_score >= 45:
        return 'D (2 points)', total_score
    else:
        return 'F (0 points)', total_score

# Function to generate grade-based recommendations using Gemini API
def generate_grade_recommendations(grade):
    if grade >= 70:
        grade_desc = "good grade"
    else:
        grade_desc = "bad grade"

    input_prompt = f"The student has a {grade_desc}. What recommendations do you have for them?"
    model = genai.GenerativeModel('gemini-pro')
    response = model.generate_content(input_prompt, generation_config=genai.types.GenerationConfig(max_output_tokens=400))
    return response.text

# Create the Streamlit app
def main():
    # Set page title and configure layout
    st.set_page_config(page_title="Exam Score Prediction", layout="wide")

    # Add a title and description
    st.title("Exam Score Prediction")
    st.markdown(
        "This app predicts exam scores based on input features such as level, course units, attendance, mid-semester score, and assignments."
    )

    # Create input fields
    col1, col2 = st.columns(2)
    with col1:
        level = st.number_input("Level", min_value=200, max_value=400, step=1)
        course_units = st.number_input("Course Units", min_value=1, max_value=4, step=1)
    with col2:
        attendance = st.slider("Attendance", min_value=1, max_value=10, step=1)
        mid_semester = st.slider("Mid Semester Score", min_value=1, max_value=20, step=1)
        assignments = st.slider("Assignments", min_value=1, max_value=10, step=1)

    # Make prediction
    if st.button("Predict Exam Score"):
        # Create input data
        input_data = np.array([[level, course_units, attendance, mid_semester, assignments]])

        # Make predictions using both models
        rf_prediction = make_rf_prediction(rf_model, input_data)
        deep_prediction = make_deep_prediction(deep_model, input_data)

        # Combine predictions
        combined_prediction = (rf_prediction + deep_prediction) / 2

        # Calculate total score
        total_score = attendance + mid_semester + assignments + combined_prediction[0]

        # Ensure total score does not exceed 100
        total_score = min(total_score, 100)

        st.write(f"Predicted Exam Score: {combined_prediction[0]:.2f}")
        st.write(f"Total Score: {total_score:.2f}")
        
        # Calculate GPA
        gpa, numeric_score = calculate_gpa(total_score)
        st.write(f"Predicted GPA: {gpa}")

        # Generate recommendations based on GPA
        recommendations = generate_grade_recommendations(numeric_score)
        st.subheader("Recommendations Based on Grade:")
        st.write(recommendations)

if __name__ == '__main__':
    main()