jofam / rethinking Goto Github PK

It is currently advertised as a legacy function, but its functionality is broken. Specifially:

• when using the default AICc, it gives an error as AICc() cannot be found
• when using AIC, it returns columns filled with zero's
• when using add.names = TRUE, it adds a column with NA values.

In my humble opinion, this function should simply go. @rmcelreath any remarks on this?

In order to get rid of all the warnings in R CMD check, using roxygen will be a ton of help. So all help pages have to be converted to roxygen2 pages.

Trying the following code :

data(cars)
flist <- alist(
  dist ~ dnorm( mu , sigma ) ,
  mu <- a+b*speed ,
  c(a,b) ~ dnorm(0,1) , 
  sigma ~ dexp(1)
)

fit <- quap( flist , 
             start=list(a=40,b=0.1,sigma=20) , 
             data=cars )

samp <- extract.samples(fit, n = 100)
HPDI(samp)

gives :

> HPDI(samp)
    |0.89     0.89| 
-1.942856  2.685888 

Whereas one would expect :

> coda::HPDinterval(coda::as.mcmc(x), prob = 0.89)
          lower     upper
a     -1.724315  1.735476
b      2.721952  3.141751
sigma 12.080041 16.184893
attr(,"Probability")
[1] 0.89

This is due to the sapply() wrongly flattening the result returned by the coda code.

There's no need for a concat() function when paste0 suffices. It's exactly that.
Contained in the file map2stan.templates. Used all over the place.

There's a bunch of confusion around the depends, imports and calling non-exported functions from other packages. This needs to be cleaned up.

The generic (in file map-quap-class.r actually calls class (instead of inherits), and contains quite a lot of code. This function definition should be the method for the specific classes (or at least for "ANY") and the generic turned into a standardGeneric.

Otherwise it won't survive code review by CRAN.