penLik.EMNewton: Fits hierarchical semiparametric regression model to...
In gitlongor/hisemi: Hierarchical Semiparametric Regression of Test Statistics

Description Usage Arguments Value Note Author(s) References See Also Examples

Fits hierarchical semiparametric regression model to t-statistics

penLik.EMNewton(tstat, x, df, spar = c(10^seq(-1,8,length=30), Inf),
        nknots = n.knots(length(tstat)), starts, 
	tuning.method = c("NIC", "CV"), cv.fold = 5, pen.order=1,
	poly.degree=pen.order*2-1, optim.method =
	c("nlminb", "BFGS", "CG", "L-BFGS-B", "Nelder-Mead", "SANN", "NR"), 
        logistic.correction = TRUE, em.iter.max = 10, 
        em.beta.iter.max = 1, newton.iter.max = 1500, 
        scale.conv = 0.001, lfdr.conv = 0.001, NPLL.conv = 0.001, 
        debugging = FALSE, plotit = TRUE, ...)

`tstat`	A numeric vector t-statistics
`x`	A numeric matrix of covariates, with `nrow(x)` being `length(tstat)`
`df`	A numeric scalar or vector of degrees of freedom
`spar`	A numeric vector of smoothing parameter lambda
`nknots`	A numeric scalar of number of knots
`starts`	An optional numeric vector of starting values
`tuning.method`	Either `'NIC'` or `'CV'`, specifying the method to choose the tuning parameter `spar`
`cv.fold`	A numeric scalar of the fold for cross-validation. Ignored if `tuning.method='NIC'`.
`pen.order`	A numeric scalar of the order of derivatives of which squared integration will be used as roughness penalty.
`poly.degree`	A numeric scalar of the degree of B-splines.
`optim.method`	A character scalar specifying the method of optimization.
`logistic.correction`	A logical scalar specifying whether or not the effective number of parameters should be corrected using a logistic curve
`em.iter.max`	A numeric scalar specifying the maximum number of EM iterations. If being `Inf`, then EM algorithm is used. If being `0`, then Newton method is used. Otherwise, EM algorithm is used initially, followed by Newton method.
`em.beta.iter.max`	A numeric scalar specifying the maximum number of iterations in the maximization step for the beta parameters in the EM algorithm. If being `Inf`, the original EM is used. If being 1 or other numbers, the generalized EM algorithm is used.
`newton.iter.max`	A numeric scalar specifying the maximum number of iterations in Newton method.
`scale.conv`	A small numeric scalar specifying the convergence criterion for the scale parameter.
`lfdr.conv`	A small numeric scalar specifying the convergence criterion for the local false discovery rates.
`NPLL.conv`	A small numeric scalar specifying the convergence criretion for the negative penalized log likelhood.
`debugging`	A logical scalar. If `TRUE`, then `dump.frame` will be called whenever error occurs.
`plotit`	A locgical scalr specifying whether a plot should be generated.
`...`	Currently not used.

An list of class hisemit:

`lfdr:`	A numeric vector of local false discovery rates.
`model`	A list of `tstat`, `df` and `x`, which are the same as arguments
`scale.fact:`	A list with `scale.fact`: Scale factor `sd.ncp`: Equivalent standard deviation of noncentrality parameters `r`: A reparameterization of `scale.fact` `t.cross`: `sqrt(df(s^(2/(df+1))-1)/(1-s^(-2df/(df+1))))` where `s` is the `scale.fact`
`pi0:`	A numeric vector of mixing proportions for the central t component
`tuning:`	A list with `mean`: Mean criterion `var`: Variance of criterion across observations `grp`: Cross-validation group membership `method`: The `tuning.method` used. `final`: The minimum mean criterion
`spar:`	A list with `all`: All smoothing parameters searched `final`: The smoothing parameter used `final.idx`: The index of the final `spar`
`enp:`	A list with `raw`: Raw effective number of parameters `logistic`: Effective number of parameters after fitting logistic curve as a correction `final`: The effective nubmer of parameters in the final model `good.idx`: The index of the selected effective number of parameters
`fit:`	A list with `intercept`: The fitted intercept `covariate.idx`: The index of covariates `f.covariate`: Each additive smooth function evaluated at the covariates `f`: Fitted smoothing funciton `beta`: Estimated regression coefficients `H`: Expanded design matrix `asym.vcov`: Asymptotic variance-covariance matrix for estimated parameters
`NPLL:`	A list with `NPLL`: Negative penalized log likelihood `logLik`: Log likelihood `penalty`: Penalty term `saturated.ll`: Saturated log likelihood

When spar is too small, the results need to be treated cautiously. It is advisable to plot the results as a check.

Long Qu rtistician@gmail.com

Long Qu, Dan Nettleton, Jack Dekkers (2012) A hierarchical semiparametric model for incorporating inter-gene relationship information for analysis of genomic data. Biometrics, 68(4):1168-1177

plot.hisemit, fitted.hisemit, coef.hisemit, vcov.hisemit, residuals.hisemit, logLik.hisemit, confint.hisemit, plot.hisemit, hisemi-package, pi0-package