jamesnzeex's picture
update model save logic
acd390c
raw
history blame
9.86 kB
import json
import pickle
import requests
import geopy.distance
import numpy as np
import pandas as pd
from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.model_selection import train_test_split
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestRegressor
def get_data_data_gov(url):
query_string = url
resp = requests.get(query_string)
data = json.loads(resp.content) # convert from json to python dict
print('Number of records:', len(data.get('result').get('records')))
data = data['result']['records']
df = pd.DataFrame.from_dict(data).drop(['_id'], axis=1)
# print(df.isnull().sum())
# print(df.dtypes)
return df
def get_data_singstat_price_index(url):
query_string = url
resp = requests.get(query_string)
data = json.loads(resp.content) # convert from json to python dict
print('Number of records:', len(data.get('Data').get('row')[0].get('columns')))
df = pd.DataFrame.from_dict(data.get('Data').get('row')[0].get('columns'))
# print(df.isnull().sum())
# print(df.dtypes)
return df
def get_data_one_map(address, is_LAT_LONG_only=False):
query_string = 'https://developers.onemap.sg/commonapi/search?searchVal='+str(address)+'&returnGeom=Y&getAddrDetails=Y'
resp = requests.get(query_string)
data = json.loads(resp.content) # convert from json to python dict
if data['found'] != 0 and is_LAT_LONG_only:
data = data['results'][0]
data = (data['LATITUDE'], data['LONGITUDE'])
elif data['found'] != 0:
data = data['results'][0]
else:
data = None
return data
def distance_to_city(address):
hdb_coordinates = get_data_one_map(address, is_LAT_LONG_only = True)
return geopy.distance.great_circle((1.29293672227779, 103.852585580366), hdb_coordinates).km
def distance_to_nearest_MRT_station(address):
hdb_coordinates = get_data_one_map(address, is_LAT_LONG_only = True)
df_MRT = pd.read_pickle('./df_MRT.pkl')
MRT_coordinates = df_MRT.T.iloc[-1,:].tolist()
dist = []
for coordinates in MRT_coordinates:
dist.append(geopy.distance.great_circle(hdb_coordinates, coordinates).km)
return min(dist)
def month_to_quarter(x):
year = int(x.split('-')[0])
month = int(x.split('-')[1])
if month <= 3:
month = '1Q'
elif month <= 6:
month = '2Q'
elif month <= 9:
month = '3Q'
else:
month = '4Q'
return (str(year) + '-' + str(month))
def get_update(data_year):
### DATA EXTRACTION AND PREPROCESSING ###
df_raw_data = get_data_data_gov('https://data.gov.sg/api/action/datastore_search?resource_id=f1765b54-a209-4718-8d38-a39237f502b3&limit=1000000')
df_raw_data['address'] = df_raw_data['block'] + ' ' + df_raw_data['street_name']
df_raw_data['quarter'] = df_raw_data['month'].apply(month_to_quarter)
df_raw_data['year_sold'] = df_raw_data['month'].apply((lambda x: int(x.split('-')[0])))
df_raw_data['month_sold'] = df_raw_data['month'].apply((lambda x: int(x.split('-')[1])))
df_raw_data['remaining_lease_years'] = df_raw_data['remaining_lease'].apply(lambda x: int(x.split()[0]))
df_raw_data['floor_area_sqft'] = round((df_raw_data['floor_area_sqm'].astype(float))*10.764)
df_price_index = get_data_singstat_price_index('https://tablebuilder.singstat.gov.sg/api/table/tabledata/M212161?isTestApi=true')
df_price_index = df_price_index.rename(columns = {'key':'quarter', 'value': 'index'})
df_price_index['quarter'] = df_price_index['quarter'].apply(lambda x : x.replace(' ', '-'))
df_selected_year = df_raw_data[df_raw_data['year_sold']>=data_year]
quarter_list = list(df_price_index['quarter'])
df_selected_year['quarter'] = df_selected_year['quarter'].apply(lambda x : x if x in quarter_list else quarter_list[-1])
df_selected_year = pd.merge(df_selected_year, df_price_index, how='left', on='quarter')
# convert to float
df_selected_year['index'] = pd.to_numeric(df_selected_year['index'])
df_selected_year['resale_price'] = pd.to_numeric(df_selected_year['resale_price'])
# normalised to latest price index
df_selected_year['normalised_resale_price'] = round(df_selected_year['resale_price']*float(df_price_index.tail(1)['index'])/df_selected_year['index'],0)
df_selected_year['price_psf'] = round(df_selected_year['normalised_resale_price']/df_selected_year['floor_area_sqft'])
df_selected_year['storey_range'] = df_selected_year['storey_range'] \
.str.replace('01 TO 03', 'Low Floor') \
.str.replace('04 TO 06', 'Mid Floor') \
.str.replace('07 TO 09', 'Mid Floor') \
.str.replace('10 TO 12', 'High Floor') \
.str.replace('13 TO 15', 'High Floor') \
.str.replace('16 TO 18', 'High Floor') \
.str.replace('19 TO 21', 'High Floor') \
.str.replace('22 TO 24', 'High Floor') \
.str.replace('25 TO 27', 'High Floor') \
.str.replace('25 TO 27', 'High Floor') \
.str.replace('25 TO 27', 'High Floor') \
.str.replace('28 TO 30', 'High Floor') \
.str.replace('31 TO 33', 'High Floor') \
.str.replace('34 TO 36', 'High Floor') \
.str.replace('37 TO 39', 'High Floor') \
.str.replace('40 TO 42', 'High Floor') \
.str.replace('43 TO 45', 'High Floor') \
.str.replace('46 TO 48', 'High Floor') \
.str.replace('49 TO 51', 'High Floor')
df_selected_year = df_selected_year.drop(columns=['street_name', 'resale_price', 'remaining_lease', 'lease_commence_date', 'block'])
HDB_address_list = df_selected_year['address'].unique().tolist()
data_list = []
for i in range(0, len(HDB_address_list)):
data = get_data_one_map(HDB_address_list[i])
if data is not None:
data_list.append(data)
df_HDB = pd.DataFrame.from_dict(data_list)
#creating primary key for df_HDB for further table join
df_HDB['LAT_LONG']= (df_HDB['LATITUDE'] +' '+ df_HDB['LONGITUDE']).apply(lambda x: tuple(x.split(' ')))
tmp_df = pd.read_csv('./MRT_LRT_STATION.csv')
MRT_list = tmp_df['Station'].unique().tolist()
print('Number of MRT stations:', len(MRT_list))
data_list = []
for i in range(0, len(MRT_list)):
data = get_data_one_map(MRT_list[i])
if data is not None:
data_list.append(data)
df_MRT = pd.DataFrame.from_dict(data_list)
df_MRT['LAT_LONG'] = (df_MRT['LATITUDE'] +' '+ df_MRT['LONGITUDE']).apply(lambda x: tuple(x.split(' ')))
df_MRT.to_pickle('df_MRT.pkl')
MRT_coordinates = df_MRT.T.iloc[-1,:].tolist()
df_HDB_coordinates = pd.DataFrame(df_HDB['LAT_LONG'])
df_HDB_coordinates = pd.DataFrame(df_HDB['LAT_LONG'])
for coordinates in MRT_coordinates:
df_HDB_coordinates[coordinates]=df_HDB['LAT_LONG'].apply(lambda y: geopy.distance.great_circle(y, coordinates).km)
# get the distance from each address to the nearest station
df_HDB_coordinates['distance_to_nearest_MRT_station'] = df_HDB_coordinates.iloc[:,1:].apply(lambda x: min(x), axis=1)
df_HDB_coordinates_with_MRT_distance = df_HDB_coordinates.iloc[:,[0,-1]]
df_HDB = pd.merge(df_HDB, df_HDB_coordinates_with_MRT_distance, on='LAT_LONG', how='left')
df_HDB = df_HDB.drop(columns=['SEARCHVAL', 'BUILDING', 'ADDRESS' ,'X', 'Y', 'LONGTITUDE'])
# City Hall: 1.29293672227779 103.852585580366
df_HDB['distance_to_city'] = df_HDB['LAT_LONG'].apply(lambda x: geopy.distance.great_circle((1.29293672227779, 103.852585580366), x).km)
df_HDB['address'] = df_HDB ['BLK_NO'] + ' ' + df_HDB ['ROAD_NAME']
df_HDB['address'] = df_HDB['address'] \
.str.replace('AVENUE', 'AVE') \
.str.replace('CRESCENT', 'CRES') \
.str.replace('ROAD', 'RD') \
.str.replace('STREET', 'ST') \
.str.replace('CENTRAL', 'CTRL') \
.str.replace('HEIGHTS', 'HTS') \
.str.replace('TERRACE', 'TER') \
.str.replace('JALAN', 'JLN') \
.str.replace('DRIVE', 'DR') \
.str.replace('PLACE', 'PL') \
.str.replace('CLOSE', 'CL') \
.str.replace('PARK', 'PK') \
.str.replace('GARDENS', 'GDNS') \
.str.replace('NORTH', 'NTH') \
.str.replace('SOUTH', 'STH') \
.str.replace('BUKIT', 'BT') \
.str.replace('UPPER', 'UPP}') \
.str.replace('COMMONWEALTH', "C'WEALTH")
df_clean_data = pd.merge(df_selected_year, df_HDB, on='address', how='left')
df_clean_data = df_clean_data.dropna(subset=['LAT_LONG'])
### FEATURE SELECTION ###
features = [
'flat_type',
'storey_range',
'floor_area_sqft',
'remaining_lease_years',
'distance_to_nearest_MRT_station',
'distance_to_city',
'price_psf']
df = df_clean_data[features]
df = pd.get_dummies(df)
### MODEL TRAINING AND TESTING ###
X = df.drop('price_psf', axis=1)
y = df['price_psf']
print('Average price_per_sqm:', y.mean())
print('Min price_per_sqm:', y.min())
print('Max price_per_sqm:', y.max())
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
model_RFR = make_pipeline(StandardScaler(),RandomForestRegressor())
model_RFR.fit(X_train, y_train) # apply scaling on training data
model_RFR.score(X_test, y_test)
print('Mean Absolute Error:', mean_absolute_error(y_test, model_RFR.predict(X_test)))
print('Mean Squared Error:', mean_squared_error(y_test, model_RFR.predict(X_test)))
print('Root Mean Squared Error:', np.sqrt(mean_squared_error(y_test, model_RFR.predict(X_test))))
f = open("model.log", "r")
data = f.readline()
model_score = float(data.split()[-1])
MSE = np.sqrt(mean_squared_error(y_test, model_RFR.predict(X_test)))
if MSE < model_score:
pickle.dump(model_RFR, open('model.sav', 'wb'))
f = open("model.log", "w")
f.write(f'model_score = {MSE}')
f.close()