2
\$\begingroup\$

I currently have the following python code that adds a few calculated columns to my consol file. Essentially it combines all the sales files into one combined DF and then adds 4 new sales columns with some conditions back to the consol file. The code works, but it takes about an hour to complete.

For reference the combined sales data probably adds up to about 2m~ rows, whilst the consol data contains 40k rows to which it is adding the sales when the criteria match. Is there anything that could be done to optimize the code to run a bit faster?

import os
import pandas as pd

# Folder paths
consol_folder = r'C:\Users\user\Desktop\consol_data'
sales_folder = r'C:\Users\user\Desktop\PowerQuery_sales'
output_folder = r'C:\Users\user\Desktop\output'
output_file = os.path.join(output_folder, 'output.xlsx')

# Check if output file already exists and remove it if it does
if os.path.exists(output_file):
    os.remove(output_file)

# Importing sales files
sales_files = os.listdir(sales_folder)
sales_data = pd.DataFrame()
for i, file in enumerate(sales_files):
    print(f"Processing sales file {i+1} of {len(sales_files)}: {file}")
    if file.endswith('.xlsx'):
        file_path = os.path.join(sales_folder, file)
        df = pd.read_excel(file_path)
        sales_data = pd.concat([sales_data, df], ignore_index=True)
print(f"Sales files imported and combined!")

# Importing consol file
consol_file = os.path.join(consol_folder, '2023_07_16 consolfile_updated.xlsx')
consol_data = pd.read_excel(consol_file)
print(f"Consol file imported!")

# Adding new columns to the consol file
consol_data['30D sales'] = 0
consol_data['60D sales'] = 0
consol_data['90D sales'] = 0
consol_data['YTD sales'] = 0

# Calculation & filters to apply 
for index, row in consol_data.iterrows():
    material_no = row['MaterialNo']
    customer_type = row['CustomerType']
    sale_group = row['SaleGroup']
    on_shelf_date = row['on-shelf date']
    c_o = row['C/O']

    if customer_type in ['FS', 'MB']:
        sales_filtered = sales_data[
            (sales_data['MaterialNo'] == material_no) &
            (sales_data['CustomerType'] == customer_type) &
            (sales_data['SaleGroup'] == sale_group) &
            (sales_data['NetSalesDate'] >= on_shelf_date)
        ]
    else:
        sales_filtered = sales_data[
            (sales_data['MaterialNo'] == material_no) &
            (sales_data['CustomerType'] == customer_type) &
            (sales_data['NetSalesDate'] >= on_shelf_date)
        ]

    consol_data.at[index, '30D sales'] = sales_filtered[
        (sales_filtered['NetSalesDate'] >= on_shelf_date) &
        (sales_filtered['NetSalesDate'] <= on_shelf_date + pd.DateOffset(days=30))
    ]['Invoiced/Sold/Shipped Qty'].sum()

    consol_data.at[index, '60D sales'] = sales_filtered[
        (sales_filtered['NetSalesDate'] >= on_shelf_date) &
        (sales_filtered['NetSalesDate'] <= on_shelf_date + pd.DateOffset(days=60))
    ]['Invoiced/Sold/Shipped Qty'].sum()

    consol_data.at[index, '90D sales'] = sales_filtered[
        (sales_filtered['NetSalesDate'] >= on_shelf_date) &
        (sales_filtered['NetSalesDate'] <= on_shelf_date + pd.DateOffset(days=90))
    ]['Invoiced/Sold/Shipped Qty'].sum()

    # Calculate YTD sales
    next_on_shelf_date = consol_data.loc[
        (consol_data['MaterialNo'] == material_no) &
        (consol_data['on-shelf date'] > on_shelf_date),
        'on-shelf date'
    ].min()

    if pd.isnull(next_on_shelf_date):
        consol_data.at[index, 'YTD sales'] = sales_filtered['Invoiced/Sold/Shipped Qty'].sum()
    else:
        if c_o == "C/O":
            consol_data.at[index, 'YTD sales'] = sales_filtered[
                (sales_filtered['NetSalesDate'] >= on_shelf_date) &
                (sales_filtered['NetSalesDate'] < next_on_shelf_date)
            ]['Invoiced/Sold/Shipped Qty'].sum()
        else:
            consol_data.at[index, 'YTD sales'] = sales_filtered['Invoiced/Sold/Shipped Qty'].sum()

    print(f"Processed row {index+1} of {len(consol_data)}")

# Save the result to a new Excel file
consol_data.to_excel(output_file, index=False)

print("Processing complete. Output file saved")
\$\endgroup\$

1 Answer 1

Reset to default
2
\$\begingroup\$

Strongly consider using pathlib.Path for your files and directories. This will also simplify your remove and your extension check via glob.

Don't call concat within the loop; only call it after the loop is done.

When adding your new columns, use list-index assignment style to assign them all at once.

Consider adopting logging to get configurable formats and levels, etc.

Up to this point, the suggestions above could look like (untested)

import logging
from pathlib import Path

import pandas as pd

desktop = Path('~/Desktop').expanduser()
consol_folder = desktop / 'consol_data'
sales_folder = desktop / 'PowerQuery_sales'
output_folder = desktop / 'output'
output_file = output_folder / 'output.xlsx'

output_file.unlink(missing_ok=True)

sales_frames = []

for i, file in enumerate(sales_folder.glob('*.xlsx')):
    logging.info('Processing sales file %d: %s', i, file)
    df = pd.read_excel(file)
    sales_frames.append(df)

sales_data = pd.concat(sales_frames, ignore_index=True)
logging.info('%d sales files imported and combined', len(sales_frames))

consol_file = consol_folder / '2023_07_16 consolfile_updated.xlsx'
consol_data = pd.read_excel(consol_file)
logging.info('Consol file imported')

consol_data[['30D sales', '60D sales', '90D sales', 'YTD sales']] = 0

In your big loop after, you need to vectorise, and the loop needs to go away. For instance, since your row comes from consol_data and you're doing comparison filters to sales_data, that's actually a join, something like (also untested)

sales_filtered = pd.join(
    left=consol_data[consol_data['CustomerType'].isin(['FS', 'MB'])],
    right=sales_data,
    on=['MaterialNo', 'CustomerType', 'SaleGroup', 'NetSalesDate'],
)

I can't be any more specific because no sample data have been provided.

\$\endgroup\$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.