How to get a lift table for a binary classification predictor

[saddas]

I developed a workflow for a binary classification model, and I would like to get information about the lift. The Lift Chart I added shows the individual and cumulative lift at different depths in the data (when sorted by highest probability). Is there a way to get to the actual table that generates the Lift Chart? I would like to know the lift value at each depth as it’s difficult to read the information accurately directly from the chart.
I am aware that the H2O Scorer provides a lift table, but is there a way to get such a table from the traditional predictor nodes?

[mlauber71]

@saddas one way to do it would be to edit the parameter table in this example and add more steps:

Create a table that counts TP, FP, TN, FN at different thresholds

Thank you all! @mlauber71 : I like these solutions. I also managed to develop a custom solution for this. It’s a longer workflow and not as elegant but gives a lot of flexibility. [image]

[saddas]

@mlauber71 : Yes, I thought about that. But I am still new to KNIME, I am not sure how to do it. My two challenges are:

1. How to aggregate by decile (10% chunks in the data). The workflow I generated calculates TP, FP, TN, FN at each cutoff point, not at each decile. I coded a variable based on rules for cutoff and then iterated across that variable.

-----------UPDATE: I now know how to address this point: Use the Chunk Loop Start node.---------

2. I am not sure how to introduce the cumulative calculation into the loop. The table should include individual lift and also cumulative lift. If I lop across each 10% of the data, how doe I keep a cumulative count in my loop?

Thanks.

[saddas]

So, essentially, what I am trying to do is this: take the first row and do some calculation, then add the second row and do a calculation on the first two rows (not just the second row), then add in the third row (and do a calculation on the first 3 rows), etc.
A chunk loop start won’t work for these specific calculations.

[saddas]

I solved this by adding an interval loop; similar to this post:
Looping over increasing partition of input data - KNIME Analytics Platform - KNIME Community Forum