Distance matrix

Dear,

 

Is it possible to write the full matrix in the distance matrix calculator node? instead of only to pass the triangular matrix? maybe an options box to chose between options ; triangular or full and not passing throught distance writer and then reader.

 

thanks

fab 

May I ask what the use case of this would be? The information is exactly the same but it needs twice as much space.

I known that is the double space and for this reason if is it possible to have this option as check box. Some investigation use the full matrix after other dimensional scaling and spatial statistics for visualize in a 2D map similarity , and one can add other properities in the map this is usefull to undestand the behavior of your space.

We could make a simple conversion node?

Distance Matrix to ListCells ?

Hi,

One reason for the full matrix instead of the triangular one is for asymmetric distance functions. For example if you calculate the sequence similarity for a protein alignment the length of the sequence goes into the calculation. This means that if the sequences have different length you get different numbers.

At the moment I haven't implemented a node for clustering protein sequences but I would need a possibility to export the full matrix before I start.

Cheers

Guido 

I thought that could be simple to have the chance to pass trought the calculator node triangular or full matrix this for not complicate always the workflow but if a conversion node in listscell or not (one can use split collection column node) is the praticable way ...is ok   

Yes, but you can already create complete distance matrix functions for assymetric distances, the framework allows this. However, currently KNIME does not provide any assymetric distance functions.

I'm a bit reluctant to add such an option to the distance matrix calculator, because every algorithm that looks at the column as an actual distance matrix (and not just a matrix) should be able to handle symmetric triangular matrices. Thus an explicit conversion into a matrix as suggested by Iris seems more appropriate to me.

If there is a particular tratment of distance matrix and an explicit conersion is the way i think that could be ok