Hands-on experience of skrub, scikit-learn and skore.
This repository contains a draft of python file, that can be converted in to a notebook, to treat the use-case of Stallion. The data comes from this fictional situation described in a kaggle competition.
The script contains already some elements, and we invite you to complete it.
Thanks to jupytext, transform the python files into notebooks, with the following command:
jupytext --to notebook stallion_volume_forecasting.pyAll the necessary libraries are listed in the requirements.
Based on the data from previous months, we want to predict the sales for the following one (testing is for Jan 2018 only), for each agency and SKU.
Country Beeristan, a high potential market, accounts for nearly 10% of Stallion & Co.’s global beer sales. Stallion & Co. has a large portfolio of products distributed to retailers through wholesalers (agencies). There are thousands of unique wholesaler-SKU/products combinations. In order to plan its production and distribution as well as help wholesalers with their planning, it is important for Stallion & Co. to have an accurate estimate of demand at SKU level for each wholesaler.
Currently demand is estimated by sales executives, who generally have a “feel” for the market and predict the net effect of forces of supply, demand and other external factors based on past experience. The more experienced a sales exec is in a particular market, the better a job he does at estimating. Joshua, the new Head of S&OP for Stallion & Co. just took an analytics course and realized he can do the forecasts in a much more effective way. He approaches you, the best data scientist at Stallion, to transform the exercise of demand forecasting.