Do we need more than 2 cycles #55
Unanswered
XingerTang
asked this question in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
AlphaImpute2 version: commit f9e40bd
Cycle here refers to the input option
cyclesUnlike AlphaPeel, AlphaImpute2 results converge really fast.
combinedmodes, there is no difference between the 1st cycle result and the 5th cycle result, except for the decrease in the segregation probability Gen2 accuracy as the number of cycles increases, which is related to issue Possble improvements in segregation outputs #54.ped_only, more cycles do improve the output accuracy. But the difference between the 1st cycle result and the 5th cycle result is small, and the difference between the 2nd cycle result and the 5th cycle result is negligible.pop_onlyis irrelevant here because it is controlled by then_phasing_cycles.The following are the results from the accuracy tests by running 5 cycles:
Results for 1 cycle:
Results for 2 cycles:
This does not only apply to the accuracy test example, for the combined imputation run for the example dataset in the
example/folder, the peeling times are suspiciously fast and similar for the imputation cycle 2 onwards. Here are the logs for the combined imputation:And
We can see the times are all
0.01seconds for peel down, and0.02seconds for peel up.The same applies to the pedigree-only imputation:
Beta Was this translation helpful? Give feedback.
All reactions