I am trying to get a weighted density estimation of some data. Unfortunately, the data is approximately exponentially distributed after the weight transform, which gives a false sense of positive density in $\mathbb{R}^-$ and a non-existing maximum in the vicinity of the origin, when using standard KDE methods.

I found this very nice answer using logsplines for the density estimation of an exponentially distributed RV. The result is impressive. Unfortunately, applying this to weighted density estimation does not seem to be straight forward.

Does logsplines support weighted density estimation in some form? Different approaches also warmly welcomed.

  • $\begingroup$ I don't think the logspline function in the logspline package has weights. $\endgroup$ – Glen_b Nov 20 '14 at 9:41
  • $\begingroup$ @Glen_b, yes, seems so, unfortunately $\endgroup$ – cel Nov 20 '14 at 10:29
  • 1
    $\begingroup$ I am just wondering why you would even have to use a density estimator for an exponential distribution. $\endgroup$ – Michael R. Chernick May 26 '17 at 20:04