Asynch-SGBDT: Asynchronous Parallel Stochastic Gradient Boosting Decision Tree based on Parameters Server

摘要: In AI research and industry, machine learning is the most widely used tool. One of important algorithms Gradient Boosting Decision Tree, i.e. GBDT whose training process needs considerable computational resources time. To shorten time, many works tried to apply on Parameter Server. However, those are synchronous parallel which fail make full use this paper, we examine possibility using asynchronous methods train model name algorithm as asynch-SGBDT (asynchronous stochastic gradient boosting decision tree). Our theoretical experimental results indicate that scalability influenced by sample diversity datasets, sampling rate, step length setting tree. Experimental also show reaches a linear speedup in manner when datasets trees meet high requirements.