I would need some help with my new idea of workflow.
Since now, I was testing my data with different H2O models. My workflow was designed to predict if patients were going to have an illness or not.
In this picture, it is shown how I splited my data with the H2O Partitioning node in 75% for training (red) and 25% for testing (blue). The training dataset is aimed to optimse the parameters for each model. The remaining testing data got results for the accuracy and other statistical parameters to prove whether the model is good or not to predict my output. Now, I would like to predict the output of a new patient (new row) which I do not know the result. Do you know how can I do it?
@helfortuny You would have to store your model in a file (maybe best a MOJO file) and then score new data with the H2O prediction node. If you use MOJO you can just use the data as they are in KNIME (given they have the same structure as the training data). If you want to use the native H2O models you will have to load them into an H2O context (like with the training data):
Can you elaborate on what you want to do. In the above example you have the option to score new data just like you scored the test data in your example.
You can evaluate the results with the Binary Classification Inspector, giving the real value (column) and the column with the scores (typically from 0-1).
You will have to define a cutoff of the score based on the results on the test data and your business case (what would a wrong classification mean/cost). You can then have your (new) prediction. The ‘truth’ will have to be observed in real life. You will not be able to further deduct the result from the new data.
