As noted by a user on Stack Overflow, we don't support plotting the ROC curve directly from R or Python.
Let's add an `h2o.plot_roc()` function in R which takes either a model or a performance object. So it could be: `h2o.plot_roc(object, train = FALSE, valid = FALSE, xval = FALSE)` which would default to printing the training ROC (since it always exists) and the object could be a H2OBinomialMetrics object or a H2OBinomialModel object. As part of this ticket, we may also decide to add an R equivalent to the roc method in Python.
In Python, we have a similar method located here for the model: `h2o.model.H2OBinomialModel.plot_roc(train=False, valid=False, xval=False)` method in Python (or should it go here? h2o.model.binomial.H2OBinomialModel.plot_roc()`) and here for the performance: `h2o.model.metrics_base.H2OBinomialModelMetrics.plot_roc()`