A parallel implementation of reinforced learning model used in analyzing risky decision making
Vinay B. Gavirangaswamy, Ajay Kumar Gupta, Aakash N. Gupta · 2016
Analyzing datasets for Risky Decision Making (RDM) is a challenging task involving the identification of varied decision making patterns and the categorization of individuals. Researchers from various fields as diverse as psychology and marketing are actively working to identify suitable techniques, which will allow understanding decision making processes better. Researchers have commonly used machine learning algorithms to model decision making processes. However, the high computational costs of most machine learning algorithms make such endeavors challenging for increasingly large datasets. One of the most promising approaches is to use ensemble clustering for RDM analysis. Ensemble clustering is computationally intensive and thus we propose to improve its performance. Our study reveals that computational overhead is introduced through the use of dimensions in ensemble cluster RDM analyses. Improving performance requires more than the parallelization of individual clustering techniques of the ensemble. We therefore propose a FIFO queue based implementation for analyzing RDM datasets using a HPC cluster on a distributed system. Our technique is able to achieve almost a linear speedup (e.g. 44.79x using 48 MPI threads). Possible shortcomings of the proposed method, opportunities for future work, and alternative parallelization scenarios are also discussed in this paper.