Error from NaN values

[pdeffebach]

Sometimes there is numerical instability leading to NaN values, and I get this error

ERROR: ArgumentError: quantiles are undefined in presence of NaNs
Stacktrace:
 [1] _quantilesort!(::Array{Float64,1}, ::Bool, ::Float64, ::Float64) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/Statistics/src/Statistics.jl:960

My understanding is that quantreg in R skips over these NaNs and prints a warning. Could QuantileRegressions do the same?

[pdeffebach]

Perhaps we can just let NaN filter through? I think the error is coming from non-convergence. In which case we should just return NaN for everything

[pkofod]

Do you have a complete stacktrace? I don't know where this would happen.

[andrewmcgr]

It comes from qreg_vcov, the line:
iqre = quantile(resid, .75) - quantile(resid, .25)