Future work: linear binning
1d
count, sum
median
2d
robust regression
nd
Future work:
compute standard errors / bootstrap standard errors?
infrastructure for passing multiple z
Kernel smoothing plus binned summary leads to many common statistics: density =~ bin + smooth, loess =~ mean + smooth, rqss =~ quantile + smooth
deal with missing values
smooth needs to create complete grid when factor = TRUE
Think about input data structure: sparse grid, represented as a coordinate list. Binned grid class = integer vector + width/origin/nbins (0 = NA). Most transformations break the grid, in which case all you case preserve is min, max and number of bins. All smoothing methods adapted to work in terms of these integers. Need to extract out bin/unbin into own class (initialised with std::vector of bin sizes)
Possible that more performance is available by switching to a sparse tensor library.
Standard errors + cut offs
Peel: implement nd version using depth
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.