I think that we could increase the speed of radEmu by using C++ for some of the computation. Calculations where this may be helpful are:
- the inner loop of
micro_fisher
- each time we calculate the log-likelihood and its derivative, this happens in
micro_fisher, macro_fisher_null, get_score_stat, f_info, and update_z
- the calls to
lapply within macro_fisher_null, especially matrix inversions, really most of computation in macro_fisher_null (if we are still using this null approach)
pseudo_huber_median- portions of
get_score_stat - specifically looping over score contributions, covariance matrix computation, and inverting I * H^T
f_info, calculation of fisher information matrix
To come back to when @adw96 has a new statdivlab RA in the fall :)
I think that we could increase the speed of
radEmuby using C++ for some of the computation. Calculations where this may be helpful are:micro_fishermicro_fisher,macro_fisher_null,get_score_stat,f_info, andupdate_zlapplywithinmacro_fisher_null, especially matrix inversions, really most of computation inmacro_fisher_null(if we are still using this null approach)pseudo_huber_medianget_score_stat- specifically looping over score contributions, covariance matrix computation, and inverting I * H^Tf_info, calculation of fisher information matrixTo come back to when @adw96 has a new statdivlab RA in the fall :)