Beac has published in the public domain PyTorch-Lifestream – a library of algorithms for constructing vector representations of event data. The library allows you to process terabyte volumes of data such as bank transactions, gaming events and purchase histories.
The library is aimed at preparing data for the subsequent training of machine learning models.
PyTorch-Lifestream contains five self-learning methods. Four of them were adapted in Assembly for event sequences, and one – contrast training – was fully developed at Sber AI Lab. Vector representations obtained in PyTorch-Lifestream will not contain personal data.
The BEAC has authorized the use of the library for research and commercial purposes.
