Hi. First, thank you for this wonderful software! I believe I found a timing issue with the convergence check in the susie-inf code.
Currently, the algorithm checks for convergence AFTER updating the variance components in preparation for the next iteration
- calculate PIP, mu, etc. using sigma/tausq from iteration i
- update sigmasq/tausq to prepare for the i+1 iteration
- check for convergence based upon PIP values from the i and i-1 iteration
- return results
We should switch step (2) and (3) above so, if we converge at iteration i, we return the variance component values (and infinitesimal posterior means) from the ith iteration.
Best,
Alex
Hi. First, thank you for this wonderful software! I believe I found a timing issue with the convergence check in the susie-inf code.
Currently, the algorithm checks for convergence AFTER updating the variance components in preparation for the next iteration
We should switch step (2) and (3) above so, if we converge at iteration i, we return the variance component values (and infinitesimal posterior means) from the ith iteration.
Best,
Alex