{"id":333207,"date":"2025-02-19T09:51:49","date_gmt":"2025-02-19T09:51:49","guid":{"rendered":"https:\/\/www.artefact.com\/?post_type=blog&p=333207"},"modified":"2025-02-27T14:30:53","modified_gmt":"2025-02-27T14:30:53","slug":"assortment-optimization-with-discrete-choice-models-in-python","status":"publish","type":"blog","link":"https:\/\/www.artefact.com\/nl\/blog\/assortment-optimization-with-discrete-choice-models-in-python\/","title":{"rendered":"Assortimentoptimalisatie met discrete keuzemodellen in Python"},"content":{"rendered":"

Author<\/h2><\/div>\"Image\"

Vincent Auriau<\/h3><\/div>

Machine Learning PhD Candidate at Artefact<\/p>\n<\/div><\/div><\/div><\/div><\/div>

<\/span><\/a><\/span>

Read our article on<\/u><\/p>\n<\/div>

<\/a><\/span><\/div>
<\/div><\/div><\/div><\/div><\/div>

Assortment optimization is a critical process in retail that involves\u00a0curating the ideal mix of products<\/strong>\u00a0to meet consumer demand while taking into account the many logistics constraints involved. The retailers need to make sure that they offer the right products, in the right quantities, at the right time. By leveraging data and consumer insights, retailers can make informed decisions on which items to stock, how to manage inventory, and what products to prioritize based on customer preferences, seasonal trends, and sales patterns.<\/p>\n

For retail businesses, assortment optimization is essential to striking a balance between\u00a0variety<\/strong>\u00a0and\u00a0efficiency<\/strong>. Offering too few choices may drive customers away, while offering too many can lead to confusion, excess inventory, and lower profit margins. Optimizing the product assortment helps businesses enhance customer satisfaction by ensuring popular items are available while eliminating underperforming products that take up valuable shelf space.<\/p>\n

Choice modeling<\/strong>\u00a0is an efficient way to approach assortment optimization because it provides a data-driven framework for understanding customer preferences and predicting how they will choose between different products. By analyzing various factors such as price sensitivity, product features, and brand loyalty, choice modeling helps retailers identify which products are most likely to meet customer demand.<\/p>\n<\/div>

Ultimately, choice modeling enables retailers to offer the right mix of products, tailor assortments to specific customer segments, and can also optimize shelf space to drive profitability or even the pricing of items.<\/p>\n

If you have never heard of choice modeling, you can read\u00a0our article<\/a>\u00a0that introduces the key concepts with examples. In this article, we will mainly focus on how discrete choice models can be used to optimize an assortment of products. We provide code samples based on the\u00a0choice-learn<\/a>\u00a0library, which is designed to help data scientists on such use cases.<\/p>\n

The provided code uses the choice-learn Python package and can be found in a notebook\u00a0<\/em>here<\/em><\/strong><\/a>.<\/em><\/p>\n<\/div>

Set up: Installing Python & Choice-Learn<\/h2><\/div>

In this article, we provide code snippets to accompany the explanations. The code uses the\u00a0Choice-Learn<\/a>\u00a0library, which provides efficient tools for choice modeling and several applications \u2014 such as assortment optimization or price. Choice-Learn is available through PyPI, you can get it simply with<\/p>\n

<\/pre>\n<\/div>
pip install choice-learn<\/div>\n
<\/pre>\n<\/div>
The dataset : sales receipts<\/h2><\/div>
We will use the TaFeng grocery dataset. You can download it from\u00a0Kaggle<\/a>\u00a0and open it in your Python environment with choice-learn:<\/p>\n
from choice_learn.datasets import load_tafeng<\/div>\n
tafeng_df = load_tafeng(as_frame=True)
\nprint(tafeng_df.head())<\/div>\n<\/div>
<\/a><\/span><\/div>
The dataset consists of over 800,000 individual purchases in a Chinese grocery store. For each purchase, various details are provided, including the purchased item (PRODUCT_ID), the price at which it was sold (SALES_PRICE), and the customer\u2019s age group (AGE_GROUP).<\/p>\n
You can observe that many different items are provided and some of them are seldom sold. In order to streamline logistics, the retailer may choose to reduce the number of products they offer. The goal in this case is to identify the optimal subset of items to sell.<\/p>\n
To achieve this, we focus on the top-selling items, as they are more likely to be purchased again and will play a crucial role in shaping a more efficient and profitable assortment.\u00a0Note that we do this mainly to simplify the example and that all items could be kept.<\/em><\/p>\n
# Keep only top 20 selling products<\/span>
\ntafeng_df = tafeng_df.loc[
\ntafeng_df.PRODUCT_ID.isin(tafeng_df.PRODUCT_ID.value_counts().index[:20<\/span>])
\n].reset_index(drop=True<\/span>)<\/div>\n
# Remove NaN values<\/span>
\ntafeng_df = tafeng_df.loc[
\ntafeng_df.AGE_GROUP.isin([“25-29”<\/span>, “40-44”<\/span>, “45-49”<\/span>, “>65”<\/span>, “30-34”<\/span>, “35-39”<\/span>, “50-54”<\/span>, “55-59”<\/span>, “60-64”<\/span>]\n)
\n].reset_index(drop=True<\/span>)<\/div>\n
print<\/span>(tafeng_df.head())<\/div>\n<\/div>
Let\u2019s also encode the age categories with one hot values every ten year:<\/p>\n
# Encoding the age categories<\/span>
\ntafeng_df[“twenties”<\/span>] = tafeng_df.apply(lambda<\/span> row: 1<\/span> if<\/span> row[“AGE_GROUP”<\/span>] == “25-29”<\/span> else<\/span> 0<\/span>, axis=1<\/span>)
\ntafeng_df[“thirties”<\/span>] = tafeng_df.apply(
\nlambda<\/span> row: 1<\/span> if<\/span> row[“AGE_GROUP”<\/span>] in<\/span> ([“30-34”<\/span>, “35-39”<\/span>]) else<\/span> 0<\/span>, axis=1<\/span>
\n)
\ntafeng_df[“forties”<\/span>] = tafeng_df.apply(
\nlambda<\/span> row: 1<\/span> if<\/span> row[“AGE_GROUP”<\/span>] in<\/span> ([“40-44”<\/span>, “45-49”<\/span>]) else<\/span> 0<\/span>, axis=1<\/span>
\n)
\ntafeng_df[“fifties”<\/span>] = tafeng_df.apply(
\nlambda<\/span> row: 1<\/span> if<\/span> row[“AGE_GROUP”<\/span>] in<\/span> ([“50-54”<\/span>, “55-59”<\/span>]) else<\/span> 0<\/span>, axis=1<\/span>
\n)
\ntafeng_df[“sixties_and_above”<\/span>] = tafeng_df.apply(
\nlambda<\/span> row: 1<\/span> if<\/span> row[“AGE_GROUP”<\/span>] in<\/span> ([“60-64”<\/span>, “>65”<\/span>]) else<\/span> 0<\/span>, axis=1<\/span>
\n)<\/div>\n<\/div>
Now that our data is ready, we need to create a\u00a0ChoiceDataset<\/strong>, the data handler object in\u00a0choice-learn<\/strong>. This involves specifying the features that describe the context in which a purchase is made:<\/p>\n