Statistics and Value Counter nodes don't seem to be domain aware

Dear KNIMErs,

is there a specific reason why the Statistics and the Value Counter nodes don't seem to be domain aware?

When working on a filtered table subset, I would expect them both to take into account the original domain of the data for their ouput, while they seem to recalculate/redefine the domain of the input table like they were preceded by a Domain Calculator node (which is not the case).

For example, if I filter out from the Iris dataset all rows of Species = "versicolor" using a Row Filter node, the domain of the output table still contains a reference to "versicolor" for the Species column, which is what I expect.

However when this filtered table is passed to a Value Counter node to count the species, only the two remaining ones are listed in the output, while, based on the original domain, I would expect also "versicolor" to be listed with count = 0.

Similarly for the Statistics node, where, based on the original domain, I would expect "versicolor" to appear in both the histogram and the occurrence table with frequency 0.

Perhaps you could add an option in the configuration dialogue of these two nodes to make them domain aware or not.

Closely related question, is there an easy way to tell which nodes are domain aware and which ones are not without having to test them out? The node documentation doesn't seem to specify this aspect.



I'm not 100% sure but I believe that no node except view nodes respects the incoming domain. They all look on the actual data.

Thanks Thorsten, that explains it.

If you ask me, at least the basic statistics nodes should have a configuration option to take into account the domain for their task and output (mostly for categorical variables). A bit like the Histogram and Histogram (Interactive) nodes that have a "Show Empty Bin" option to show zero occurency bins.