Setting trans to 2 and K > 1 leads to error in flowClust

[HVinther]

I installed flowClust using pak::pkg_install. The reported version is 3.46.0.

Here is a quick reproducible example,

set.seed(42)

df<-data.frame(
  "x" = runif(100),
  "y" = runif(100)
)

fc<-flowClust(df, trans = 2, K = 2)

With the traceback,

The error seems to come from this snippet

flowClust/R/flowClust.R

Lines 803 to 806 in 24c0ffd

	prior$order<-obj$oorder;
	if(trans==1&obj$lambda==1){
	obj$mu<-rbox(obj$mu,obj$lambda)
	}

Perhaps the condition should be

all(c(trans==1,obj$lambda==1))

or something similar?

[HVinther]

A shorter example would be to use the bundled data,

data(rituximab)
flowClust::flowClust(rituximab, K = 2, trans = 2)

which yields the same error message.

[HVinther]

Having made the change proposed above, the flowClust algorithm run without errors. The resulting object, however, has loglikelihood, BIC etc. as NAN, as seen in the summary below:

I installed flowClust using pak::pkg_install. The reported version is 3.46.0.

Here is a quick reproducible example,

set.seed(42)

df<-data.frame(
  "x" = runif(100),
  "y" = runif(100)
)

fc<-flowClust(df, trans = 2, K = 2)

With the traceback,

The error seems to come from this snippet

flowClust/R/flowClust.R

Lines 803 to 806 in 24c0ffd
prior$order<-obj$oorder;
if(trans==1&obj$lambda==1){
obj$mu<-rbox(obj$mu,obj$lambda)
}

Perhaps the condition should be

all(c(trans==1,obj$lambda==1))

or something similar?

[HVinther]

Looking through the scripts in the testthat directory there seems to be no tests with trans = 2, which seems rather odd. Is this option not really supported, even though it is mentioned in both the documentation and the article, https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-10-145 ?

[gfinak]

Probably a bug that is related to changes in how R handles vector arguments in if statements many releases ago.
Tells you something about how much use cluster-specific transformation gets in flowClust.

[HVinther]

Any idea on what might lead to the missing log likelihood and similar estimates? I have a bit of a hard time figuring out the out code with the current formatting.