It would be great to extend the method to clamped about bspline-regression

[rstebbing]

Hi @rsbandaru,

Apologies for the delayed response! Just to clarify, what do you mean by "missing"? With regards to README-2.png, for example, are you referring to the fact that, for the leftmost and rightmost input data points, the model has not extended far enough so that the error between the model and these data points is perpendicular?

[christianbrugger]

I think this is what is meant:

Cubic spline not clamped. This is how it is implemented:

Clamped cubic splines. Here the curve touches the first and last control point. This is usually how it is implemented in most application and also shown on Wikipedia.

The difference is the knots vector.

knots_unclamped = [0, 0.167, 0.333, 0.5, 0.667, 0.833, 1]
knots_clamped = [0, 0, 0, 0.5, 1, 1, 1]

Here is the website I took the screenshots from:

http://nurbscalculator.in/

I agree with rsbandaru that it would be good to have this feature.

I am not sure if this can easily be done, as I read you assume uniform knot spacing. In this case an alternative could be to add the first and last points multiple times (depending on the degree). However it seems to give a slightly different curve. And it would require some special casing in the optimizer, to treat them always the same.