dealing with scaler range near 0

[tremblap]

As discovered by Rodrigo, when ranges are near zero, the algorythm gets unstable. In line with ScikitLearn, this is a proposed sanitizer of all ranges that might become denominators in scaling.

[tremblap]

test code in SC for fun:

b = Buffer.sendCollection(s, 8.collect{|i|7.collect{|j|i+(j/10)}.scramble}.flat,7 )
b = Buffer.sendCollection(s, 8.collect{|i|7.collect{|j|i+(j/10)}}.flat,7 )
b = Buffer.sendCollection(s, 8.collect{|i|7.collect{|j|i}}.flat,7 )

c = FluidDataSet(s)

c.fromBuffer(b,1)
c.read("/Users/pa/Desktop/fluidobject.json")

c.print

d = FluidDataSet(s)

e = FluidRobustScale(s)

e.fitTransform(c,d)

d.print


f = FluidNormalize(s)

f.fitTransform(c,d)

d.print


g = FluidStandardize(s)

g.fitTransform(c,d)

d.print


===
h = Buffer.sendCollection(s, 8.collect{1.0.rrand(-1)})
h = Buffer.sendCollection(s, 8.collect{|i|i/4 - 0.75}) //[ -0.75, -0.5, -0.25, 0.0, 0.25, 0.5, 0.75, 1.0 ]
i = Buffer(s)
j = Buffer(s)

e.transformPoint(h,i,{i.getn(0,8,{|a|a.postln})})
e.inverseTransformPoint(i,j,{j.getn(0,8,{|a|a.postln})})

f.transformPoint(h,i,{i.getn(0,8,{|a|a.postln})})
f.inverseTransformPoint(i,j,{j.getn(0,8,{|a|a.postln})})

g.transformPoint(h,i,{i.getn(0,8,{|a|a.postln})})
g.inverseTransformPoint(i,j,{j.getn(0,8,{|a|a.postln})})

j.getn(0,8,{|a|a.postln})

s.dumpOSC