Abstract:
A variety of Machine Learning (ML) frameworks are being instigated to let developers build ML algorithms easily. The study include an evaluation of various Al toolkits that includes Tensor Flow, Scikit-Learn and Keras. We evaluated the frameworks using theoretical and empirical parameters where theoretical parameters are languages, fault tolerance, ease of debugging, ONNX support, mobile support, multiple GPU support and style of programming and empirical parameters are training time, accuracy, precision, recall and fl score . We used an approach where we kept a control environment and kept same settings of frameworks on all experiments to get the best results. An analytical table discussed in section 5.1 shows the compatibility of the framework against the specific parameters used for theoretical evaluation and for empirical evaluation , tables are shown in section 5 that shows the performance measures of the frameworks being evaluated.
