Palantir results not exactly reproducible
Marius1311 opened this issue · 4 comments
Marius1311 commented
Setting the random seed does not lead to exactly reproducible results, you can easily check this in an example, just set the random seed, run palantir twice and compare the actual fate probabilities numerically, there will be slight deviations. This can only be eliminated when the waypoint cells are manually cached (written to file) and provided to palantir. In that case, results are reproducible. The deviations are really small and don't lead to e.g. different gene expression trends, however, it would still be nice to fix this bug. Thanks!
wangjiawen2013 commented
+1
katosh commented
This should be resolved with f4970f9. It will also become available with Palantir v1.2.
wangjiawen2013 commented
Great, I'll give it a try
Marius1311 commented
Amazing, thanks!