compute_ephemera | R Documentation |
compute_ephemera()
does data-independent pre-computations for sped and can speed up repeated applications
compute_ephemera(gtwid, hn, padding, spline_dim, perknot = 2)
gtwid |
Object of class spedecon_gtwid describing the density of |
hn |
Object of class |
padding |
Support of spline space is extended by |
spline_dim |
Numeric integer, dimension of spline space |
perknot |
Number of positivity constraints per knot |
The computations in sped rely on several matrices and vectors that are determined by the error density, spline space, and histogram bins, but do not depend on the data. Computing these is the most time-intensive element of the process, so if the estimator will be applied several times to different data, but the same error density, spline space, and histogram bins (likely in simulations), gains can be had by pre-computing those matrices and vectors just one time.
For comparison, the sped function internally uses padding = 0.4
, and perknot = 2
.
Object of class spedecon_ephemera
, a list containing the pre-computed values.
Kent D, Ruppert D (2023). “Smoothness-Penalized Deconvolution (SPeD) of a Density Estimate.” Journal of the American Statistical Association, to appear. ISSN 0162-1459, \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1080/01621459.2023.2259028")}
alpha <- 1e-3; n <- 1e3; s <- 0.3
Y <- rgamma(n,5,2) + rnorm(n,0,s)
gtwid <- gaussian_gtwid(sd=s)
hn <- hist(Y,breaks="FD",plot=FALSE)
ephemera <- compute_ephemera(gtwid=gtwid,hn=hn,padding=0.4,spline_dim=30,perknot=2)
sol1 <- sped(Y,gtwid,1e-3,ephemera=ephemera) # fast
sol2 <- sped(Y,gtwid,1e-3) # slow
attr(sol1,"coef") - attr(sol2,"coef")
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.