PSMART: parameter server based multiple additive regression trees system
Proceedings of the 26th International Conference on World Wide Web Companion, 2017•dl.acm.org
… To this end, in this work, we introduce PSMART, a Parameter Server (PS) based
distributed learning system that reliably scales MART to support hundreds of billions of
samples and thousands of features in the clusters. Our contributions are: (1) We built a
general MART system based on PS which enjoys efficient sparse communication, better
fault tolerance and ease of use (like the SQL style) etc; (2) We implemented a distributed
LambdaMART with a query id based partitioning strategy. Due to these techniques, the …
distributed learning system that reliably scales MART to support hundreds of billions of
samples and thousands of features in the clusters. Our contributions are: (1) We built a
general MART system based on PS which enjoys efficient sparse communication, better
fault tolerance and ease of use (like the SQL style) etc; (2) We implemented a distributed
LambdaMART with a query id based partitioning strategy. Due to these techniques, the …
In this paper, we describe a Parameter Server based Multiple Additive Regression Trees system, or PSMART for short. Empirically, PSMART scales MART to hundreds of billions of samples and thousands of features in production.
ACM Digital LibraryShowing the best result for this search. See all results
