h2o-tutorial | H2O Tuning and Ensembling Tutorial for R | Machine Learning library

It looks like you are missing the parameter keep_cross_validation_predictions=True in each of your models. For example you would want to do the following for your GLM and then likewise for the other models you want to stack:

Try this, it'll do your job:

please see the documentation for the full parameter list. For your convenience here is the list confusion_matrix(metrics=None, thresholds=None, train=False, valid=False, xval=False).

Here is a working example of how to use the method:

Starting with your 2nd question, Flow has a precision/recall curve (and it is interactive). Flow is always running on port 54321 of each node, i.e. http://127.0.0.1:54321 if you are running h2o locally.

I imagine that there is something interesting with your data or model, and that when you look at the precision/recall curve it will become clear.

In R if you do str(m) (where m is your model) you will see all the model data. m@training_metrics@metrics$thresholds_and_metric_scores$recall holds the recall numbers for each threshold.

I cannot work out how to look inside the Python object, yet, but your call was correct. On my quick test (the iris dataset with a 2-category enum column added):