I would like to use a smooth version of binned data (the shape is roughtly a “Lorentzian + constant”). I noticed that the class TKDE could be used for that. Is there a way to provide binned data to the TKDE object?
It doesn’t really make sense to provide binned data to a KDE. A KDE typically places a kernel (usually a Gaussian) centered at each data point. The smoothed distribution is then constructed from the sum of these kernels. By providing binned data you limit the final resolution of the smooth distribution. Is there a problem with providing the full data set?