app.py

import streamlit as st
import pandas as pd
from io import BytesIO
from itertools import product
from statsmodels.tsa.statespace.sarimax import SARIMAX
import plotly.express as px

st.set_page_config(layout="wide")

# Function to run the SARIMAX Model
def run_sarimax(city_data, order, seasonal_order):
    try:
        model = SARIMAX(city_data, order=order, seasonal_order=seasonal_order, enforce_stationarity=False, enforce_invertibility=False)
        model_fit = model.fit(disp=False)
        return model_fit.forecast(steps=6), model_fit.aic
    except Exception as e:
        st.error(f"An error occurred during model fitting: {e}")
        return None, None

def create_data():
    # Assuming you have a CSV file named 'accident_count.csv' with 'City' and 'Accident Count' columns
    data = pd.read_csv('accident_count.csv', parse_dates=True, index_col=0)
    data.index = pd.to_datetime(data.index, format='%Y%m')
    data = data.groupby('City').resample('M').sum().reset_index()
    data.index = data['Accident Month Bracket']
    data = data.drop(['Accident Month Bracket'], axis=1)
    data.index = data.index.strftime('%Y-%m')
    return data

def to_excel(df):
    output = BytesIO()
    writer = pd.ExcelWriter(output, engine='xlsxwriter')
    df.to_excel(writer, sheet_name='Sheet1')
    writer.save()
    processed_data = output.getvalue()
    return processed_data

# Initialize session state for best parameters
if 'best_params' not in st.session_state:
    st.session_state.best_params = {'order': (1, 1, 1), 'seasonal_order': (1, 1, 1, 12)}

st.title("SARIMAX Forecasting")

# Data preparation
data = create_data()
unique_cities = data['City'].unique()

# Creating tabs for each city
tabs = st.tabs([city for city in unique_cities])

for tab, city in zip(tabs, unique_cities):
    with tab:
        # SARIMAX specific sliders
        p = st.slider('AR Order (p)', 0, 5, value=st.session_state.best_params['order'][0], key=city+'p')
        d = st.slider('Differencing Order (d)', 0, 2, value=st.session_state.best_params['order'][1], key=city+'d')
        q = st.slider('MA Order (q)', 0, 5, value=st.session_state.best_params['order'][2], key=city+'q')
        P = st.slider('Seasonal AR Order (P)', 0, 5, value=st.session_state.best_params['seasonal_order'][0], key=city+'P')
        D = st.slider('Seasonal Differencing Order (D)', 0, 2, value=st.session_state.best_params['seasonal_order'][1], key=city+'D')
        Q = st.slider('Seasonal MA Order (Q)', 0, 5, value=st.session_state.best_params['seasonal_order'][2], key=city+'Q')
        S = st.slider('Seasonal Period (S)', 1, 24, value=st.session_state.best_params['seasonal_order'][3], key=city+'S')

        city_data = data[data['City'] == city]['Accident Count']
        forecast, aic = run_sarimax(city_data, (p, d, q), (P, D, Q, S))

        if forecast is not None:
            st.write(f"Best Parameters with AIC: {aic}")
            st.write(f"Non-Seasonal Order: {(p, d, q)}, Seasonal Order: {(P, D, Q, S)}")
            forecast_index = pd.date_range(start=city_data.index[-1], periods=7, freq='M')[1:]
            forecast_index = forecast_index.to_period('M')  # Convert to period index with monthly frequency
            forecast_df = pd.DataFrame(forecast, columns=['Forecast'])
            forecast_df = forecast_df.round(0)
            st.table(forecast_df)
            fig = px.line(forecast_df, x=forecast_df.index, y="Forecast")
            st.plotly_chart(fig)

            # Grid search button
            if st.button(f'Run Grid Search for {city}'):
                best_aic = float('inf')
                best_params = None
                # Define the range for each parameter
                p_range = d_range = q_range = range(3)
                P_range = D_range = Q_range = range(3)
                S = 12  # Assuming a fixed seasonal period, adjust as needed
            
                # Perform the grid search
                for params in product(p_range, d_range, q_range, P_range, D_range, Q_range):
                    order = params[:3]
                    seasonal_order = params[3:] + (S,)
                    try:
                        _, temp_aic = run_sarimax(city_data, order, seasonal_order)
                        if temp_aic < best_aic:
                            best_aic = temp_aic
                            best_params = (order, seasonal_order)
                    except Exception as e:
                        st.error(f"An error occurred for parameters {params}: {e}")
            
                # Update the session state with the best parameters
                if best_params is not None:
                    st.session_state.best_params = {
                        'order': best_params[0],
                        'seasonal_order': best_params[1]
                    }
                    st.write(f"Best Parameters for {city}: {best_params} with AIC: {best_aic}")


        # Export to Excel button
        if st.button(f'Export {city} to Excel'):
            df_to_export = forecast_df
            excel_data = to_excel(df_to_export)
            st.download_button(label='📥 Download Excel', data=excel_data, file_name=f'{city}_forecast.xlsx', mime='application/vnd.ms-excel')

# Rest of your code