Design and Implementation of Machine Learning Evaluation Metrics on HPCC Systems

The HPCC Systems Production Machine Learning bundles provide a diverse set of features that allow the parallelized creation and training of Machine learning models and a large set of evaluation metrics that can be used to test the trained model to ascertain its performance. To help monitor the models more closely however, a new set of evaluation methods that incorporate the analysis of clusters and the selection of features, as well as other commonly used tests, have been proposed, implemented, and tested. The implementations are written completely in Enterprise Control Language and support the various features provided by the Machine Learning bundles such as the Myriad Interface. This paper provides a comprehensive summary of the evaluation metrics currently available in the library, before presenting the details of the design and implementation of the new evaluation methods. It the goes on to present the results of testing these implementations against implementations present in the python scikit-learn library, and a few data visualisations demonstrating some uses of the implemented evaluation metrics.