How can I split a dataset into training and testing data and plot graphs and matrices in ml5?

Question

For a classification problem, I want to split a json dataset into training and testing data, and use the it to train the model and later test its accuracy using testing data. How can I do this in ml5? Even if I manually split the dataset into two separate files, how do I get the actual labels and compare it with the predicted labels of the testing dataset to get the accuracy of the model? I also want to plot a confusion matrix and some other graphs.By setting debug to true I can only see the loss. In ml5 documentation there is a property called neuralNetworkVis where all of the tf-vis operations are organized. But I could not find any reference or examples on it. I would like to know how to use it. Or is there something that I can use?