The input data (same code for each format) calculates the center[0] & bounds-scale for all points when loading InputData.
I've just done a refactor to remove this, as in some previous changes it seemed like removing this broke it...
But now [that I've done my changes properly] it seems to be more than happy training in original space.
Is there something I'm missing, or was this just here to auto-center/shrink the output for say, a viewer/tool, and not anything to do with the training?
Or does it effect the training in some way that I can't see?
[0] It's actually the average, not the center :) (means != center of bounds)
The input data (same code for each format) calculates the center[0] & bounds-scale for all points when loading
InputData.I've just done a refactor to remove this, as in some previous changes it seemed like removing this broke it...
But now [that I've done my changes properly] it seems to be more than happy training in original space.
Is there something I'm missing, or was this just here to auto-center/shrink the output for say, a viewer/tool, and not anything to do with the training?
Or does it effect the training in some way that I can't see?
[0] It's actually the average, not the center :) (means != center of bounds)