Rescale case weights the way TDA's cwt(wnorm=) does

tda_cwt_norm reproduces TDA's cwt(wnorm=s)=W; (or the bare-flag cwt(wnorm)=W;) as a plain vector transform, for weights= on tda_qreg, tda_glm, tda_ple, tda_ltb and tda_rate – none of which expose wnorm as an argument, since this is all it does: fitted coefficients come back identical whether the weights are rescaled this way first or not, only the standard errors change (rescaling is a constant multiplier on the whole weighted log-likelihood, so it moves the curvature at the optimum, not where the optimum is). cwt(wnorm=s)=W; on the raw weights and plain cwt=W; on tda_cwt_norm(W, s) give the same fit, to the last printed digit.

Usage

tda_cwt_norm(w, s = length(w))

Arguments

w

a numeric vector of raw case weights.

s

the target sum for the rescaled weights – TDA's wnorm=s. Left at its default, length(w) (the number of cases), it matches the bare-flag form, cwt(wnorm)=W;, TDA's "keep the same effective sample size" correction.

Value

w, rescaled so it sums to s.

Examples

w <- c(2, 2, 2, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10)  # sums to 74
tda_cwt_norm(w)         # cwt(wnorm)=W;    -- rescaled to sum 13 (length(w))
#>  [1] 0.245283 0.245283 0.245283 1.226415 1.226415 1.226415 1.226415 1.226415
#>  [9] 1.226415 1.226415 1.226415 1.226415 1.226415
tda_cwt_norm(w, 100)    # cwt(wnorm=100)=W;
#>  [1] 1.886792 1.886792 1.886792 9.433962 9.433962 9.433962 9.433962 9.433962
#>  [9] 9.433962 9.433962 9.433962 9.433962 9.433962