Is the integer cast in abs() really intended?

[sofievg]

Is the integer cast in abs() really intended?

[SamGG]

Interesting point.
I don't think there is such a stop condition in the implementation of the kohonen package.
https://github.com/rwehrens/kohonen/blob/28eb3599bdc423e94a4a1e09310ffd9de667ead8/kohonen/src/supersom.cpp#L93-L173

[SofieVG]

Dear @LTLA,

This integer cast was indeed not intended and has already been changed to fabs() on my local installation. Apparently that update hasn't made it to this github yet, but I will update soon.

The idea behind the change parameter is indeed to check for convergence, but I agree that the value 1 can be kind of arbitrary, especially when the package is getting wider use. I think we choose this value back then because it made sense for the datasets we were working with at the time. I'll consider updating this in the future, either at least make it a parameter, or indeed, as you are suggesting, computing it based on the data range itself.

Kind regards,
Sofie

[LTLA]

Okay, good to see that it's fixed.