""" Get Florida cannabis lab results | Flowery Copyright (c) 2023-2024 Cannlytics Authors: Keegan Skeate Created: 5/18/2023 Updated: 4/28/2024 License: Description: Archive Florida cannabis lab result data for the Flowery. Data Sources: - [The Flowery](https://support.theflowery.co) """ # Standard imports: from datetime import datetime import os from time import sleep # External imports: from cannlytics.data.web import initialize_selenium from cannlytics.utils.constants import DEFAULT_HEADERS import pandas as pd import requests # Selenium imports. from selenium.webdriver.common.by import By from selenium.webdriver.support import expected_conditions as EC def get_results_the_flowery( data_dir: str, slug = 'the-flowery', producer_license_number = 'MMTC-2019-0020', lists_url = 'https://support.theflowery.co/hc/en-us/sections/7240468576283-Drop-Information', overwrite = False, ): """Get lab results published by The Flowery on the public web.""" # Initialize the web driver. driver = initialize_selenium() # Load the lists page to get each list of COAs. coa_lists = [] driver.get(lists_url) links = driver.find_elements(by=By.TAG_NAME, value='a') for link in links: if 'COAs' in link.text: coa_lists.append(link.get_attribute('href')) # Get COA URLs. coa_urls = [] for coa_list in coa_lists: driver = initialize_selenium() driver.get(coa_list) links = driver.find_elements(by=By.TAG_NAME, value='a') for link in links: href = link.get_attribute('href') if href and href.endswith('.pdf'): coa_urls.append(href) driver.close() driver.quit() # Close the browser. driver.close() driver.quit() # Create an output directory. datasets_dir = os.path.join(data_dir, '.datasets') if not os.path.exists(datasets_dir): os.makedirs(datasets_dir) # Save the COA URLs. date = datetime.now().isoformat()[:19].replace(':', '-') df = pd.DataFrame(coa_urls) df.to_excel(f'{datasets_dir}/fl-lab-result-urls-{slug}-{date}.xlsx', index=False) print('Saved %i lab result URLs for %s' % (len(df), slug)) # Create a directory for COA PDFs. pdf_dir = os.path.join(datasets_dir, 'pdfs') if not os.path.exists(pdf_dir): os.makedirs(pdf_dir) # Create a directory for each licensees COAs. license_pdf_dir = os.path.join(pdf_dir, producer_license_number) if not os.path.exists(license_pdf_dir): os.makedirs(license_pdf_dir) # Download the COA PDFs. for coa_url in coa_urls: sample_id = coa_url.split('/')[-1].split('.')[0] batch_id = coa_url.split('/')[-2] outfile = os.path.join(license_pdf_dir, f'{batch_id}-{sample_id}.pdf') if os.path.exists(outfile) and not overwrite: continue sleep(0.3) response = requests.get(coa_url, headers=DEFAULT_HEADERS) with open(outfile, 'wb') as pdf: pdf.write(response.content) print('Downloaded: %s' % outfile) # TODO: Save details about each COA, including The Flowery company data. # Return the COA URLs. return df def get_product_results_the_flowery( data_dir: str, overwrite = False, **kwargs, ): """Get product results from The Flowery website.""" # Initialize a web driver. driver = initialize_selenium() # Iterate over all of the product types. observations = [] categories = ['Flower', 'Concentrates', 'Pre-Rolls', 'Vaporizers', 'Tinctures'] for category in categories: # Load the category page. url = f'https://theflowery.co/shop?categories[]={category}' driver.get(url) sleep(3.33) # Get all of the product cards. divs = driver.find_elements(by=By.CSS_SELECTOR, value='a.s-shop-product-card') for div in divs: # Extract product name. product_name = div.find_element(by=By.CLASS_NAME, value='title').text # Extract product image URL. product_image = div.find_element(by=By.TAG_NAME, value='img').get_attribute('src') # Extract product price. product_price = float(div.find_element(by=By.CLASS_NAME, value='full-price').text.strip('$')) # Extract strain type. strain_type = div.find_element(by=By.CLASS_NAME, value='sort').text # Extract product URL (assuming the URL is stored in the href attribute of a link) product_url = div.get_attribute('href') # Store the extracted data in a dictionary obs = { 'product_name': product_name, 'image_url': product_image, 'product_price': product_price, 'strain_type': strain_type, 'product_url': product_url } observations.append(obs) # Get the COA URL for each product. # TODO: Also get the image_url. for i, obs in enumerate(observations): coa_url = '' driver.get(obs['product_url']) sleep(3.33) links = driver.find_elements(by=By.TAG_NAME, value='a') for link in links: if link.get_attribute('href') and '.pdf' in link.get_attribute('href'): coa_url = link.get_attribute('href') break observations[i]['coa_url'] = coa_url print('Found COA URL: %s' % coa_url) # Close the driver. driver.close() # Download the COA PDFs. license_pdf_dir = os.path.join(data_dir, '.datasets', 'pdfs', 'MMTC-2019-0020') for obs in observations: coa_url = obs['coa_url'] # Get the sample ID sample_id = coa_url.split('/')[-1].split('.')[0] # Format the file. outfile = os.path.join(license_pdf_dir, f'{sample_id}.pdf') if os.path.exists(outfile) and not overwrite: print('Cached: %s' % outfile) continue sleep(0.3) # Download the PDF. # FIXME: This is failing every time. # try: response = requests.get(coa_url, headers=DEFAULT_HEADERS) with open(outfile, 'wb') as pdf: pdf.write(response.content) print('Downloaded: %s' % outfile) sleep(3.33) # except: # print('Failed to download: %s' % coa_url) # Merge The Flowery data with the COA data. the_flowery = {'business_dba_name': 'The Flowery'} observations = [{**the_flowery, **x} for x in observations] # Save the data. date = datetime.now().isoformat()[:19].replace(':', '-') data = pd.DataFrame(observations) data.to_excel(f'{data_dir}/the-flowery-lab-result-urls-{date}.xlsx', index=False) print('Saved %i lab result URLs for The Flowery.' % len(data)) return data # === Test === # [✓] Tested: 2024-04-14 by Keegan Skeate if __name__ == '__main__': # Specify where your data lives. DATA_DIR = 'D://data/florida/results' # [✓] TEST: Get The Flowery COAs. try: the_flowery_products = get_product_results_the_flowery(DATA_DIR) except Exception as e: print('ERROR:', e) try: the_flowery_coas = get_results_the_flowery(DATA_DIR) except Exception as e: print('ERROR:', e)