Make smoothing window a flag

[martintoreilly]

Smoothing only works if the underlying x-axis is a unit "grid" and then the window must be 1. This should be a boolean flag where the name makes it clear what it is doing and the type of data it can be applied to.

e.g. something like diffuse_to_unit_interval?

[andeElliott]

I presume this is for NetEMD, if so then I agree :-)

[martintoreilly]

I think there are two potential robustness / generalisable approaches to smoothing.

• Generic binning (may or may not merge discrete locations)
  • @andeElliott This should be easy with the existing C++ code
• Convolutional smoothing (sliding window / soft binning)
  • Both @andeElliott and @leospinaf like this but we are all worried about making unnecessary extra work before publication. @leospinaf thinks it is important for an option / custom code to reproduce paper

[andeElliott]

One proposal that came up in a discussion with (@leospinaf) that is simple but gets us quite a way there.

Provide a discretisation function, that transforms the variables, prior to being send to the NetEmd comparison/optimisation with four implemented options (and let the user define others if needed)

1 Identity function (nothing is done) 2 Round to nearest integer. 3 Discretise to midpoint of user provided breaks.

4. Simple histogram binning approach, i.e. make bins with breaks at minValue+x*(maxValue-minValue)/10 for x=0,...,10

We then leave the complex stuff for later work.

What do you think? @martintoreilly @leospinaf