cannabis_licenses / algorithms /get_licenses_me.py
Keegan Skeate
🍅NJ + 🎴NV + 🦞ME + 🌲WA licenses | Updated 🗺️
825c643
raw
history blame
6.24 kB
"""
Cannabis Licenses | Get Maine Licenses
Copyright (c) 2022 Cannlytics
Authors:
Keegan Skeate <https://github.com/keeganskeate>
Candace O'Sullivan-Sutherland <https://github.com/candy-o>
Created: 9/29/2022
Updated: 9/30/2022
License: <https://github.com/cannlytics/cannlytics/blob/main/LICENSE>
Description:
Collect Maine cannabis license data.
Data Source:
- Maine Office of Cannabis Policy
URL: <https://www.maine.gov/dafs/ocp/open-data/adult-use>
"""
# Standard imports.
from datetime import datetime
import os
from typing import Optional
# External imports.
from bs4 import BeautifulSoup
from cannlytics.data.gis import geocode_addresses
from dotenv import dotenv_values
import pandas as pd
import requests
# Specify where your data lives.
DATA_DIR = '../data/me'
ENV_FILE = '../.env'
# Specify state-specific constants.
STATE = 'ME'
MAINE = {
'licensing_authority_id': 'MEOCP',
'licensing_authority': 'Maine Office of Cannabis Policy',
'licenses': {
'url': 'https://www.maine.gov/dafs/ocp/open-data/adult-use',
'key': 'Adult_Use_Establishments_And_Contacts',
'columns': {
'LICENSE': 'license_number',
'LICENSE_CATEGORY': 'license_type',
'LICENSE_TYPE': 'license_designation',
'LICENSE_NAME': 'business_legal_name',
'DBA': 'business_dba_name',
'LICENSE_STATUS': 'license_status',
'LICENSE_CITY': 'premise_city',
'WEBSITE': 'business_website',
'CONTACT_NAME': 'business_owner_name',
'CONTACT_TYPE': 'contact_type',
'CONTACT_CITY': 'contact_city',
'CONTACT_DESCRIPTION': 'contact_description',
},
}
}
def get_licenses_me(
data_dir: Optional[str] = None,
env_file: Optional[str] = '.env',
):
"""Get Maine cannabis license data."""
# Create the necessary directories.
file_dir = f'{data_dir}/.datasets'
if not os.path.exists(data_dir): os.makedirs(data_dir)
if not os.path.exists(file_dir): os.makedirs(file_dir)
# Get the download link.
licenses_url = None
licenses_key = MAINE['licenses']['key']
url = MAINE['licenses']['url']
response = requests.get(url)
soup = BeautifulSoup(response.content, 'html.parser')
links = soup.find_all('a')
for link in links:
try:
href = link['href']
except KeyError:
continue
if licenses_key in href:
licenses_url = href
break
# Download the licenses workbook.
filename = licenses_url.split('/')[-1]
licenses_source_file = os.path.join(file_dir, filename)
response = requests.get(licenses_url)
with open(licenses_source_file, 'wb') as doc:
doc.write(response.content)
# Extract the data from the license workbook.
licenses = pd.read_excel(licenses_source_file)
licenses.rename(columns=MAINE['licenses']['columns'], inplace=True)
licenses['licensing_authority_id'] = MAINE['licensing_authority_id']
licenses['licensing_authority'] = MAINE['licensing_authority']
licenses['license_designation'] = 'Adult-Use'
licenses['premise_state'] = STATE
licenses['license_status_date'] = None
licenses['license_term'] = None
licenses['issue_date'] = None
licenses['expiration_date'] = None
licenses['business_structure'] = None
licenses['business_email'] = None
licenses['business_phone'] = None
licenses['activity'] = None
licenses['parcel_number'] = None
licenses['premise_street_address'] = None
licenses['id'] = licenses['license_number']
# Remove duplicates.
licenses.drop_duplicates(subset='license_number', inplace=True)
# Replace null DBA with legal name.
criterion = licenses['business_dba_name'].isnull()
licenses.loc[criterion,'business_dba_name'] = licenses['business_legal_name']
# Convert certain columns from upper case title case.
cols = ['business_legal_name', 'business_dba_name', 'business_owner_name']
for col in cols:
licenses[col] = licenses[col].apply(
lambda x: x.title().strip() if isinstance(x, str) else x
)
# Get the refreshed date.
date = licenses_source_file.split('\\')[-1].split('.')[0].replace(licenses_key, '')
date = date.replace('%20', '')
date = '-'.join([date[:2], date[2:4], date[4:]])
licenses['data_refreshed_date'] = pd.to_datetime(date).isoformat()
# Geocode licenses to get `premise_latitude` and `premise_longitude`.
config = dotenv_values(env_file)
google_maps_api_key = config['GOOGLE_MAPS_API_KEY']
cols = ['premise_city', 'premise_state']
licenses['address'] = licenses[cols].apply(
lambda row: ', '.join(row.values.astype(str)),
axis=1,
)
licenses = geocode_addresses(
licenses,
api_key=google_maps_api_key,
address_field='address',
)
drop_cols = ['state', 'state_name', 'address', 'formatted_address',
'contact_type', 'contact_city', 'contact_description']
licenses.drop(columns=drop_cols, inplace=True)
gis_cols = {
'county': 'premise_county',
'latitude': 'premise_latitude',
'longitude': 'premise_longitude',
}
licenses.rename(columns=gis_cols, inplace=True)
# Save and return the data.
if data_dir is not None:
timestamp = datetime.now().isoformat()[:19].replace(':', '-')
licenses.to_excel(f'{data_dir}/licenses-{STATE.lower()}-{timestamp}.xlsx')
return licenses
# === Test ===
if __name__ == '__main__':
# Support command line usage.
import argparse
try:
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument('--d', dest='data_dir', type=str)
arg_parser.add_argument('--data_dir', dest='data_dir', type=str)
arg_parser.add_argument('--env', dest='env_file', type=str)
args = arg_parser.parse_args()
except SystemExit:
args = {'d': DATA_DIR, 'env_file': ENV_FILE}
# Get licenses, saving them to the specified directory.
data_dir = args.get('d', args.get('data_dir'))
env_file = args.get('env_file')
data = get_licenses_me(data_dir, env_file=env_file)