bgctw / lognorm Goto Github PK

estimateSumLognormalSample currently estimates sigma parameter on the subset of finite values. But provides the correlation matrix (or autocorrelation series) for the full time series. It should use consistent parameters, i.e. the full time series with NA.
Only after constructing the correlation matrix, the non-finite cases should be filtered.

Currently, with adding a large number of lognormal random variables, a large variance-covariance matrix is constructed, e.g. from the autocorrelation components. This consumes memory and impairs performance. Rather use sparse matrix representation.

Currently, the computation of number of effective observations does not handle NA values.
If there are NA values, the return value either should be NA or omit the NA-cases.

I suggest adding argument na.rm = FALSE, as in the mean function.

Currently, estimateSumLognormal handles NAs gracefully by neglecting those terms to the sum.
This is useful for working with samples, but is prone to error if one does not expect NAs.

I suggest returning NA coefficents by default, and allow for NA by new argument na.rm = FALSE.