speed up the mode estimation procedure?

[mengyin]

The mode estimation procedure might be too slow especially when we need search the optimal mode in a big range. E.g. in Joyce's Poisson count (singlecell data) application, the counts range is (0,900) so it takes pretty long time to estimate the mode.

We guess one possible way to fix the issue is to start search with a better starting point? However, current ashn() uses stats::optimize function, which doesn't allow user to provide starting point...