pvalue plot vs uniform; FDR

qqunif  = function(p,BH=T,...)
{
  nn = length(p)
  xx =  -log10((1:nn)/(nn+1))
  plot( xx,  -sort(log10(p)),
     xlab=expression(Expected~~-log[10](italic(p))),
        ylab=expression(Observed~~-log[10](italic(p))),
       ... )
  abline(0,1,col='gray')
  if(BH)
    {
      lines( xx, -log10(0.05*(1:nn)/(nn+1)), col='red')
      lines( xx, -log10(0.10*(1:nn)/(nn+1)), col='orange')
      lines( xx, -log10(0.25*(1:nn)/(nn+1)), col='yellow')
      legend('bottomright', c("FDR = 0.05","FDR = 0.10","FDR = 0.25"),
             col=c('red','orange','yellow'),lty=1, cex=1.4)
    }
}

imagesc in CDAT

The imagesc function in MATLAB creates plots that have color bars scaled to a user-specified range. This allows you to simply create a whole series of plots that can be easily intercompared.

The following function imitates imagesc:

    import vcs
    def imagesc(x, mini, maxi):
        t = vcs.init()
        a = t.getboxfill('quick')
        a.level_1 = mini
        a.level_2 = maxi
        t.plot(x, a)

x is the input array and mini and maxi define the range in values to plot. The function establishes a VCS canvas, imports the "quick" boxfill graphics method, sets the level range to mini and maxi, and plots x using this customized graphics method. 


other tips for CDAT/Python check http://www.johnny-lin.com/cdat_tips/

likelihood ratio test

> logLik.test
function(fit0,fit)
{
  l0 = logLik(fit0)
  l1 = logLik(fit)
  lratio = abs(as.numeric(l0)-as.numeric(l1))
  df = abs( attr(l1,'df') - attr(l0,'df') )
  pval = 1-pchisq( 2*lratio, df)
  return(pval)
}

log likelihood from lm, glm, polr

> fit = lm(y ~ x)
> logLik(fit)
'log Lik.' -36.4 (df=4)
> attr(logLik(fit),'df')
[1] 4