app.py

import streamlit as st
import json
from autogluon.multimodal import MultiModalPredictor
import pandas as pd
from geopy.geocoders import GoogleV3
import os
import tempfile

st.set_page_config(layout="wide")

if "price_text" not in st.session_state:
    st.session_state.price_text = 0


@st.cache_resource
def load_geocoder():
    return GoogleV3(api_key=os.environ.get("GOOGLE_MAP_API_KEY"))


geocoder = load_geocoder()


@st.cache_resource
def load_mm_text_no_price_model():
    return MultiModalPredictor.load("models/mm-text-no-price/", verbosity=0)


mm_text_no_price_predictor = load_mm_text_no_price_model()


@st.cache_resource
def load_city_map():
    return json.load(open("city-map.json"))


city_map = load_city_map()


@st.cache_resource
def load_city_district_map():
    return json.load(open("city-district-map.json"))


city_district_map = load_city_district_map()

CERT_STATUS = pd.CategoricalDtype(
    categories=["Không có", "hợp đồng", "sổ đỏ / sổ hồng"], ordered=False
)
DIRECTION = pd.CategoricalDtype(
    categories=[
        "Không có",
        "Tây - Nam",
        "Đông - Nam",
        "Đông - Bắc",
        "Tây - Bắc",
        "Nam",
        "Tây",
        "Bắc",
        "Đông",
    ],
    ordered=False,
)
CITY = pd.CategoricalDtype(categories=city_map.keys(), ordered=False)
DISTRICT = pd.CategoricalDtype(
    categories=sum([list(map(int, v.keys())) for v in city_district_map.values()], []),
    ordered=False,
)

location_options = st.columns([1, 1, 2, 1, 1])
with location_options[0]:
    city = st.selectbox(
        "Choose city", options=city_map.items(), format_func=lambda x: x[1]
    )
with location_options[1]:
    district = st.selectbox(
        "Choose district",
        options=city_district_map[city[0]].items(),
        format_func=lambda x: x[1],
    )
with location_options[2]:
    location = st.text_input("Enter precise location")

location = (location + ", " if location else "") + city[1] + ", " + district[1]
geocode_result = geocoder.geocode(query=location, region="vn", language="vi")
latitude = geocode_result.latitude
longitude = geocode_result.longitude

with location_options[3]:
    latitude = st.number_input(
        "Enter latitude", value=latitude, step=1e-8, format="%.7f"
    )
with location_options[4]:
    longitude = st.number_input(
        "Enter longitude", value=longitude, step=1e-8, format="%.7f"
    )

numerical_options = st.columns(6)
with numerical_options[0]:
    area = st.number_input("Area (m2)", min_value=1.0)
with numerical_options[1]:
    bedrooms = st.number_input("Number of bedrooms", min_value=1, value=1)
with numerical_options[2]:
    bathrooms = st.number_input("Number of bathrooms", min_value=1, value=1)
with numerical_options[3]:
    floors = st.number_input("Number of floors", min_value=1, value=1)
with numerical_options[4]:
    front_width = st.number_input(
        "Front width, leave 0 for N/A", min_value=0.0, value=0.0, step=0.1
    )
with numerical_options[5]:
    road_width = st.number_input(
        "Road width, leave 0 for N/A", min_value=0.0, value=0.0, step=0.1
    )

cat_time_columns = st.columns(4)
with cat_time_columns[0]:
    timestamp = st.date_input("Date posted", format="DD/MM/YYYY")
with cat_time_columns[1]:
    cert_status = st.selectbox("Certification status", options=CERT_STATUS.categories)
with cat_time_columns[2]:
    direction = st.selectbox("Direction", options=DIRECTION.categories)
with cat_time_columns[3]:
    balcony_direction = st.selectbox("Balcony direction", options=DIRECTION.categories)

description = st.text_area("Description")
title = description.split(".", maxsplit=1)[0]

uploaded_image = st.file_uploader("Upload an image")
image_tmp = None
if uploaded_image:
    image_tmp = tempfile.NamedTemporaryFile(suffix=uploaded_image.name)
    image_tmp.write(uploaded_image.read())
    print(image_tmp.name)

df = pd.DataFrame(
    [
        {
            "Title": title,
            "Area": area,
            "Location": location,
            "Time stamp": timestamp,
            "Certification status": cert_status,
            "Direction": direction,
            "Bedrooms": bedrooms,
            "Bathrooms": bathrooms,
            "Front width": front_width or float("nan"),
            "Floor": floors,
            "Description": description,
            "Image URL": image_tmp.name if image_tmp else None,
            "Road width": road_width or float("nan"),
            "City_code": city[0],
            "DistrictId": int(district[0]),
            "Lattitude": latitude,
            "Longitude": longitude,
            "Balcony_Direction": balcony_direction,
        }
    ]
).astype(
    {
        "Title": "str",
        "Area": "float",
        "Location": "str",
        "Time stamp": "datetime64[ns]",
        "Certification status": CERT_STATUS,
        "Direction": DIRECTION,
        "Bedrooms": "int",
        "Bathrooms": "int",
        "Front width": "float",
        "Floor": "int",
        "Description": "str",
        "Image URL": "str",
        "Road width": "float",
        "City_code": CITY,
        "DistrictId": DISTRICT,
        "Lattitude": "float",
        "Longitude": "float",
        "Balcony_Direction": DIRECTION,
    }
)

if st.button("Get estimated price with text"):
    st.session_state.price_text = mm_text_no_price_predictor.predict(
        df, as_pandas=False
    ).item()
st.text(
    "Estimated price: {0:,} VND".format(int(st.session_state.price_text * 1e6))
    if st.session_state.price_text
    else "No price estimated."
)