----------------------------------

spline fit to logit of cdf

> xval = sort(geyser$wait)
  summary(diff(xval))   ## lots of tied observations !!
  
> tmptab = table(geyser$wait)
  xval = as.numeric(names(tmptab))  ### only 52 distinct values!

> splgey = smooth.spline(xval, -log(300/(cumsum(tmptab) - 1), 
         spar=1.e-6)
> plot(xval, plogis(predict(splgey)$y))

### This gives df, not derivative
###  the differentiated spline is very erratic

> splgey2 = smooth.spline(xval, cumsum(tmptab)/300, spar=1.e-3)
  plot(xval, predict(splgey2,deriv=1)$y)
  plot(lowess(xval,predict(splgey2,deriv=1)$y))


> splgey2 = smooth.spline(xval, cumsum(tmptab)/300, nknots=20, spar=1.e-3)
 plot(xval, predict(splgey2, deriv=1)$y)


> splgey3 = smooth.spline(xval, cumsum(tmptab)/300, nknots=10, spar=1.e-3)
 plot(xval, predict(splgey3, deriv=1)$y)


> splgey3 = smooth.spline(xval, cumsum(tmptab)/300, nknots=10, spar=1.e-3)
>  plot(xval, predict(splgey3, deriv=1)$y)
>