Pipeline Based Big Data Sketching

Big Data pipeline is an architecture for big data applications where all the sub-modules are composed in a sequential manner. In this architecture, output of one stage forwarded as an input to the next stage of the pipeline. Although, each stage of the pipeline has relevant information, but the data present at intermediate stages is not sufficient to predict inference due to uncertainty present at that stage. Therefore, a probabilistic inference based method is proposed for making strong local inference. For sketching probabilistic inference of Big Data pipeline, this work contributes probabilistic pipeline sketching of inference methods using variable elimination and bucket elimination inference.