Yandex, one of the leading technology companies that build intelligent products and services powered by machine learning, recently announced CatBoost 1.0.0, a new major version of their open-source machine learning library. The new version goes far beyond just an upgrade and is the culmination of four years of work by the Yandex Team.
CatBoost has been used by tech innovators for everything since its initial launch in 2017, from streaming service user recommendations to particle classification to destination prediction for Careem, a ride-hailing service.
The library is based on gradient boosting, a form of machine learning that analyses a wide range of data inputs by progressively training more complex models to maximize the accuracy of predictions.
“Far beyond a run-of-the-mill system update, CatBoost 1.0.0 is the transformation from an open-source machine learning library to a stand-alone, ready-to-use product. We are proud of the incredible, diligent work our team has put into reaching this milestone and are committed to continuing to improve and innovate CatBoost,” said Stanislav Kirillov, head of Yandex ML Systems Group.
This major version of CatBoost fixes all bugs and includes a number of improvements, including:
- Spark support for distributed learning.
- New, improved and convenient documentation with open code.
- Speed boost for CPU and GPU learning and overall faster model training (binary classification on CPU is 15-35% faster).
- Predictive, multiple label classification.
The major change with this release is that the Apache Spark package is now truly distributed. In its previous version, CatBoost stored test datasets in controller process memory. Now the test datasets are splitted evenly by workers. To learn more about the new version, click on the link here.
