This example demonstrates how powerful a real relational learning algorithm can be. Based on a public-domain dataset on consumer behavior, we use a propostionalization algorithm to predict whether purchases were made as a gift. We show that with relational learning, we can get an AUC of over 90%. The generated features would have been impossible to build by hand or by using brute-force approaches.
Author: Patrick Urbanke
Relational learning is one of the most underappreciated fields of machine learning. Even though relational learning is very relevant to many real world data science projects, many data scientists don't even know what relational learning is.
There are many subdomains of relational learning, but the most important one is extracting features from relational data: Most business data is relational, meaning that it is spread out over several relational tables. However, most machine learning algorithms require that the data be presented in the form of a single flat table. So we need to extract features from our relational data. Some people also call this data wrangling.
Most data scientists we know extract features from relational data manually or by using crude, brute-force approaches (randomly generate thousands of features and then do a feature selection). This is very time-consuming and does not produce good features.
The Consumer Expenditure Data Set is a public domain data set provided by the American Bureau of Labor Statistics (https://www.bls.gov/cex/pumd.htm). It includes the diary entries, where American consumers are asked to keep diaries of the products they have purchased each month.
These consumer goods are categorized using a six-digit classification system the UCC. This system is hierarchical, meaning that every digit represents an increasingly granular category.
For instance, all UCC codes beginning with ‘200’ represent beverages. UCC codes beginning with ‘20011’ represents beer and ‘200111’ represents ‘beer and ale’ and ‘200112’ represents ‘nonalcoholic beer’ (https://www.bls.gov/cex/pumd/ce_pumd_interview_diary_dictionary.xlsx).
The diaries also contain a flag that indicates whether the product was purchased as a gift. The challenge is to predict that flag using other information in the diary entries.
This can be done based on the following considerations:
In total, there are three tables which we find interesting:
ZEISS is working on delivering highly reliable machines and support processes. GetML allows building new predictive maintenance applications in a fraction of the time.
FastProp is our unique take on propositionalization. In real-world benchmarks against popular propositionalization libraries, we find FastProp is between 34x to 179x faster than the current state of the art.