app.py

import numpy as np
import pandas as pd
import geopandas as gpd
from shapely.geometry import Point, LineString
import pickle
import folium
import requests
import streamlit as st
import seaborn as sns
from streamlit_folium import st_folium
import matplotlib.pyplot as plt
from datetime import datetime
import gspread
from oauth2client.service_account import ServiceAccountCredentials
from utils import *

########## Title for the Web App ##########
st.write(client_email)
st.title("Property Price Predictor")
st.markdown('_Creator: GOH Hong Aik [[LinkedIn]](https://www.linkedin.com/in/hongaikgoh/)_')
st.markdown("""This app predicts your house price based on a few indicators, and displays amenities within 2 km. Please note the following:
- Only property types allowed are HDB, Condominium, Executive Condominium and Apartment.
- Model was trained on **resale, leasehold** properties from 2021 onwards to account for COVID-19 effects; predictions for new and/or freehold properties will not be accurate.
- Geospatial features are obtained from OneMap API with postal code. Newer postal codes may not be available (eg. property under construction)
""")

testing_postal = st.text_input('Test your postal code here')
if st.button('Click to retrieve data!'):
    result = test_postal(testing_postal)
    if result == 'INVALID LOCATION':
        st.error('Invalid postal code, you may attempt to search for the name or address instead.')
    else:
        st.success('SEARCHVAL: ' + result[0] + ' ADDRESS: ' + result[1] + ' LATITUDE: ' + str(result[2]) + ' LONGITUDE: ' + str(result[3]))

    

st.header('Predict your property price here!')

with st.form("inputs"):
    input_postal = st.text_input('Enter your postal code here e.g. 123456')
    input_storey = st.text_input('Enter your property storey here e.g. 5')
    input_age_asof_transaction = st.text_input('Enter the age in years of your property here e.g. 12')
    input_floor_area_sqft = st.text_input('Enter the floor area in square feet of your property here e.g. 756')
    input_propertyType = st.selectbox(
        'Enter your property type here',
        ('HDB', 'Condominium', 'Executive Condominium', 'Apartment'))
    propertyType_Apartment = 0
    propertyType_Condominium = 0
    propertyType_Executive_Condominium = 0
    propertyType_HDB = 0

    submitted = st.form_submit_button("Predict!")

    if submitted:
        # validate postal
        if len(input_postal) == 6:
            try:
                validate = int(input_postal)
            except:
                st.error('Invalid postal code, please enter a 6 digit integer.')
                st.stop()
        else:
            st.error('Invalid postal code, please enter a 6 digit integer.')
            st.stop()

        # validate storey
        try:
            input_storey = int(input_storey)
        except:
            st.error('Invalid storey, please enter an integer.')
            st.stop()
        
        # validate age
        try:
            input_age_asof_transaction = int(input_age_asof_transaction)
        except:
            st.error('Invalid age, please enter an integer.')
            st.stop()
        
        # validate floor area
        try:
            input_floor_area_sqft = float(input_floor_area_sqft)
        except:
            st.error('Invalid floor area, please enter a number.')
            st.stop()

        try:
            lat, long = getSVY21(input_postal)
        except ValueError:
            st.error('This postal code does not exist. Please use the widget above to test if the postal code exists.')
            st.stop()
   
        latlong_df = pd.DataFrame({'lat': [lat], 'long': [long]})
        latlong_geo = gpd.GeoDataFrame(latlong_df, crs="EPSG:4326",
                                    geometry=[Point(xy) for xy in zip(latlong_df['long'], latlong_df['lat'])])
        latlong_geo.to_crs(epsg=3857, inplace=True)
        
        # Filter geosubset to only properties (sold since 2017)
        try:
            geosubset = geosubset[(geosubset['storey'] <= input_storey + 5) & (geosubset['storey'] >= input_storey - 5)] # filter storey +- 5
            geosubset = geosubset[(geosubset['age_asof_t'] <= input_age_asof_transaction + 3) & (geosubset['age_asof_t'] >= input_age_asof_transaction - 3)] # filter age +- 3
            geosubset = geosubset[(geosubset['floor_area'] <= input_floor_area_sqft + 100) & (geosubset['floor_area'] >= input_floor_area_sqft - 100)] # filter floor area to +- 50m
            geosubset = geosubset[geosubset['propertyTy'] == input_propertyType] # filter property type
            geosubset['distance_to_property'] = geosubset.geometry.apply(lambda x: latlong_geo.distance(x)) # filter property radius within 1km
            geosubset = geosubset[geosubset['distance_to_property'] <= 1000] # filter distances within 1km
        except:
            pass

        if input_propertyType == 'Apartment':
            propertyType_Apartment = 1
        elif input_propertyType == 'Condominium':
            propertyType_Condominium = 1
        elif input_propertyType == 'Executive Condominium':
            propertyType_Executive_Condominium = 1
        elif input_propertyType == 'HDB':
            propertyType_HDB = 1        


        latlong_geo['district'] = postal_final.loc[postal_final['2dpostal'] == input_postal[:2], 'district'].item() # all districts are covered in train set
        latlong_geo['storey'] = input_storey
        latlong_geo['age_asof_transaction'] = input_age_asof_transaction
        latlong_geo['mindist_expway'] = latlong_geo.geometry.apply(lambda x: expressway.distance(x).min())
        latlong_geo['mindist_mrt'] = latlong_geo.geometry.apply(lambda x: geo_mrt.distance(x).min())
        latlong_geo['mindist_mall'] = latlong_geo.geometry.apply(lambda x: geo_mall.distance(x).min())
        latlong_geo['mindist_prisch'] = latlong_geo.geometry.apply(lambda x: geo_pri.distance(x).min())
        latlong_geo['n_prisch_2km'] = latlong_geo.geometry.apply(lambda x: (geo_pri.distance(x) < 2000).sum())
        latlong_geo['mindist_secsch'] = latlong_geo.geometry.apply(lambda x: geo_sec.distance(x).min())
        latlong_geo['propertyType_Apartment'] = propertyType_Apartment
        latlong_geo['propertyType_Condominium'] = propertyType_Condominium
        latlong_geo['propertyType_Executive_Condominium'] = propertyType_Executive_Condominium
        latlong_geo['propertyType_HDB'] = propertyType_HDB


        # Do Predictions, Convert to total price with CI (different CI for different property type)
        latlong_geo_nolatlong = latlong_geo.drop(columns=['lat', 'long', 'geometry'])
        latlong_geo_nolatlong_encoded = enc.transform(latlong_geo_nolatlong)

        prediction_psf = xgb.predict(latlong_geo_nolatlong_encoded).item()
        prediction_price = round(prediction_psf * input_floor_area_sqft)

        if latlong_geo_nolatlong_encoded['propertyType_Apartment'].item() == 1:
            st.success(f'The predicted price of your property is \\${prediction_price:,}, with 95% confidence interval at \\${int(prediction_price - 2*sd_apt*input_floor_area_sqft):,} - \\${int(prediction_price + 2*sd_apt*input_floor_area_sqft):,}.')

        elif latlong_geo_nolatlong_encoded['propertyType_Condominium'].item() == 1:
            st.success(f'The predicted price of your property is \\${prediction_price:,}, with 95% confidence interval at \\${int(prediction_price - 2*sd_condo*input_floor_area_sqft):,} - \\${int(prediction_price + 2*sd_condo*input_floor_area_sqft):,}.')

        elif latlong_geo_nolatlong_encoded['propertyType_Executive_Condominium'].item() == 1:
            st.success(f'The predicted price of your property is \\${prediction_price:,}, with 95% confidence interval at \\${int(prediction_price - 2*sd_ec*input_floor_area_sqft):,} - \\${int(prediction_price + 2*sd_ec*input_floor_area_sqft):,}.')
                
        elif latlong_geo_nolatlong_encoded['propertyType_HDB'].item() == 1:
            st.success(f'The predicted price of your property is \\${prediction_price:,}, with 95% confidence interval at \\${int(prediction_price - 2*sd_hdb*input_floor_area_sqft):,} - \\${int(prediction_price + 2*sd_hdb*input_floor_area_sqft):,}.')

        # between Jan 2017 and Oct 2022, propertyType, age, floor area range, storey
        # use geosubset to have boxplot and folium map

        if len(geosubset) > 0:
            st.markdown('There are ' + str(len(geosubset)) + ' resale ' + input_propertyType + 's transacted between Jan 2017 and Oct 2022 in your area which are similar to yours (drag the markers around as they might be overlapping):')

            map = folium.Map(location=[1.290270, 103.851959], zoom_start=11)

            folium.Circle(
                location=[latlong_geo['lat'].item(), latlong_geo['long'].item()], 
                tooltip='Your Home', 
                popup='Your Home',
                color='crimson',
                radius=10
            ).add_to(map)
    
            for index, row in geosubset.iterrows():
                
                folium.Marker(
                        [row['lat'], row['long']], 
                        tooltip=f'Resale date: {row["transactDa"].strftime("%b %Y")} <br>\n'
                        f'Storey: {int(row["storey"])} <br>\n'
                        f'Age: {int(row["age_asof_t"])} <br>\n'
                        f'Area: {int(row["floor_area"])}sqft <br>\n'
                        f'Resale Price: ${int(row["resale_pri"]):,}',
                        icon=folium.Icon(color="blue", icon='usd', prefix='fa'),
                        opacity=0.8,
                        draggable=True,
                        popup=f'Resale date: {row["transactDa"].strftime("%b %Y")} <br>\n'
                        f'Storey: {int(row["storey"])} <br>\n'
                        f'Age: {int(row["age_asof_t"])} <br>\n'
                        f'Area: {int(row["floor_area"])}sqft <br>\n'
                        f'Resale Price: ${int(row["resale_pri"]):,}'
                    ).add_to(map)

            st_folium(map, width=700, height=450)

            fig, ax = plt.subplots()
            sns.swarmplot(data=geosubset, y='resale_pri', orient='v', ax=ax)
            plt.scatter(0, prediction_price, color='orange')
            plt.text(0.05, prediction_price, 'Your predicted property price')
            plt.ylabel('Resale Price ($)')
            plt.title('Swarmplot of nearby similar properties')
            st.pyplot(fig)
        
        else:
            st.markdown('Oops! There are no resale ' + input_propertyType + 's transacted between Jan 2017 and Oct 2022 in your area which are similar to yours :(')

        # Get folium visualisations
        temp_mrt = latlong_geo.geometry.apply(lambda x: geo_mrt.distance(x)).melt().drop(columns=['variable'])
        temp_mrt = temp_mrt[temp_mrt['value'] <= 2000]
        temp_mrt = temp_mrt.merge(geo_mrt[['STN_NAME', 'lat','long']], how='left', left_index=True, right_index=True)
        temp_mrt['category'] = 'MRT'
        temp_mrt.columns = ['distance', 'name', 'lat', 'long', 'category']

        temp_mall = latlong_geo.geometry.apply(lambda x: geo_mall.distance(x)).melt().drop(columns=['variable'])
        temp_mall = temp_mall[temp_mall['value'] <= 2000]
        temp_mall = temp_mall.merge(geo_mall[['mall', 'lat','long']], how='left', left_index=True, right_index=True)
        temp_mall['category'] = 'Mall'
        temp_mall.columns = ['distance', 'name', 'lat', 'long', 'category']

        temp_pri = latlong_geo.geometry.apply(lambda x: geo_pri.distance(x)).melt().drop(columns=['variable'])
        temp_pri = temp_pri[temp_pri['value'] <= 2000]
        temp_pri = temp_pri.merge(geo_pri[['Name', 'lat','long']], how='left', left_index=True, right_index=True)
        temp_pri['category'] = 'Primary School'
        temp_pri.columns = ['distance', 'name', 'lat', 'long', 'category']

        temp_sec = latlong_geo.geometry.apply(lambda x: geo_sec.distance(x)).melt().drop(columns=['variable'])
        temp_sec = temp_sec[temp_sec['value'] <= 2000]
        temp_sec = temp_sec.merge(geo_sec[['Name', 'lat','long']], how='left', left_index=True, right_index=True)
        temp_sec['category'] = 'Secondary School'
        temp_sec.columns = ['distance', 'name', 'lat', 'long', 'category']

        temp_results = pd.concat([temp_mrt, temp_mall, temp_pri, temp_sec]).reset_index(drop=True).sort_values(['category', 'distance'])

        st.markdown('**Here are the nearest amenities to your property:**')

        for index, row in temp_results.groupby('category')[['category', 'distance', 'name']].head(2).iterrows():
        
            if row['category'] == 'MRT':
                st.markdown(row["name"] + ' :metro:: ' + str(int(row["distance"])) + 'm')
            elif row['category'] == 'Mall':
                st.markdown(row["name"] + ' :shopping_trolley:: ' + str(int(row["distance"])) + 'm')
            elif row['category'] == 'Primary School':
                st.markdown(row["name"] + ' :school:: ' + str(int(row["distance"])) + 'm')  
            elif row['category'] == 'Secondary School':
                st.markdown(row["name"] + ' :mortar_board:: ' + str(int(row["distance"])) + 'm')

        m = folium.Map(location=[1.290270, 103.851959], zoom_start=11)

        folium.Marker(
            [latlong_geo['lat'].item(), latlong_geo['long'].item()], 
            tooltip='Your Home', 
            popup='Your Home',
            icon=folium.Icon(color="orange", icon='home', prefix='fa')
        ).add_to(m)

        folium.Circle(
            radius=2000,
            location=[latlong_geo['lat'].item(), latlong_geo['long'].item()],
            color="black",
            fill=False,
            weight=1
        ).add_to(m)

        for index, row in temp_results.iterrows():
            
            if row['category'] == 'MRT':
                folium.Marker(
                    [row['lat'], row['long']], 
                    tooltip=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>',
                    icon=folium.Icon(color="red", icon='train', prefix='fa'),
                    popup=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>'
                ).add_to(m)
                
            elif row['category'] == 'Mall':
                folium.Marker(
                    [row['lat'], row['long']], 
                    tooltip=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>',
                    icon=folium.Icon(color="lightgreen", icon='shopping-cart', prefix='fa'),
                    popup=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>'
                ).add_to(m)
                
            elif row['category'] == 'Primary School':
                folium.Marker(
                    [row['lat'], row['long']], 
                    tooltip=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>',
                    icon=folium.Icon(color="lightblue", icon='book'),
                    popup=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>'
                ).add_to(m)
                
            elif row['category'] == 'Secondary School':
                folium.Marker(
                    [row['lat'], row['long']], 
                    tooltip=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>',
                    icon=folium.Icon(color="darkblue", icon='book', prefix='fa'),
                    popup=f'{row["name"]} <br>Distance: {int(row["distance"])}m</br>'
                ).add_to(m)
    
        st_folium(m, width=700, height=450)
    
        # Log searches
        keydict = {'type': service_account,
                'project_id': project_id,
                'private_key_id': private_key_id,
                'private_key': private_key,
                'client_email': client_email,
                'client_id': client_id,
                'auth_uri': auth_uri,
                'token_uri': token_uri,
                'auth_provider_x509_cert_url': auth_provider_x509_cert_url,
                'client_x509_cert_url': client_x509_cert_url}

        scope = ['https://www.googleapis.com/auth/spreadsheets',
         "https://www.googleapis.com/auth/drive"]

        credentials = ServiceAccountCredentials.from_json_keyfile_dict(keydict, scope)
        client = gspread.authorize(credentials)

        sheet = client.open("PropertyPricePredictor_Sheet").sheet1
        sheet.append_row([datetime.now().strftime('%Y-%m-%d %H:%M:%S'), int(input_postal),input_storey,input_age_asof_transaction,input_floor_area_sqft,input_propertyType])