parameter initialization discrepancy among flowClust APIs

[mikejiang]

There are two ways of using flowClust, either through flowClust directly

flowClust(fr, params, K = K)

Or through tmixFilter

 tmix_filter <- tmixFilter(params, K = K)
 filter(fr, tmix_filter)

You would expect they return the same clustering results but as @gfinak and I just found out they actually don't.

It turns out that one of the parameters is set to different values through two APIs, i.e.
randomStart is set to 0 for flowClust but 10 for tmixFilter, which leads to the entirely different behaviors.

To avoid the future confusion, I think we may set it to 10 for both since another internal API .flowClustK also uses 10.

It would be also helpful to pass most of parameters from flowClust to .flowClustK through ... to avoid the potential discrepancy.

[gfinak]

No, set it to 0, and use the kmeans initialization. The random initialization is not converging properly and needs to be debugged.

[mikejiang]

Sure

[gfinak]

I've tracked down the issue. The tol parameter needs to be decreased. Default is 1e-5, in cases where it doesn't converge after random initialization, it should be lowered, one or two orders of magnitude should be sufficient.