Possible Conditional Boxplot and Binary Classification Inspector bugs.

Hi,
I upload small workflow where I had some inconsistences in results.

  1. I have different results in the GroupBy node and Conditional boxplot for quartiles. See top part of the workflow.
  2. What means “Median Rank” and “Mean Rank” in Kruskal-Wallis Test node? That’s not means or medians, as I see.
  3. In the Binary Classification Inspector tool I have threshold equal to 0.5539999999999999. When I use it throw Rule Engine I have the same table as in Binary Classification Inspector tool. But updated table in this node (top output port) is wrong (compare scorer results). It looks like rounding error take place.
    Are this bugs or I made something wrong?
    questions_knime.knwf (90,9 КБ)

Hi Max,
Thanks for doing those tests. As for point 1: the GroupBy has different estimation methods for calculating the Quantiles available. If you choose the R_2 one in the advanced tab of the Aggregation Settings dialog, the same value as in the box plot is returned.

I cannot say anything about point 2, but for point 3 I assume that the Binary Classification Inspector uses “greater or equal”. When I use the >= operator in the Rule Engine I get the same confusion matrix as with the Binary Classification Inspector.
Kind regards,
Alexander

1 Like

Hi Alexander,

Thank you for the response.
For the point 1 I understood. I will use it in future, but for the new user it will look like a bug.
For the point 3: updated table in this node (top output port) is wrong. It’s a bug, do you agree?
Thank you, again.

Hi Max,
I don’t understand. It’s just a slight difference in how the threshold is treated. >= instead of >. I would not see this as a bug.
Kind regards
Alex

1 Like

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.