PBS #368
PBS #368
Conversation
sgkit/stats/popgen.py
Outdated
| variables.validate(ds, {stat_Fst: variables.stat_Fst_spec}) | ||
|
|
||
| fst = ds[variables.stat_Fst] | ||
| fst = fst.clip(min=0, max=0.99999) |
There was a problem hiding this comment.
0.99999 feels a bit arbitrary and I worry this will lead to numerical artefacts for values of Fst near 1. Something like
>>> 1 - np.finfo(float).epsneg
0.9999999999999999
>>> np.log(1 - np.finfo(float).epsneg)
-1.1102230246251565e-16
seems a bit more defensible.
There was a problem hiding this comment.
This looks very reasonable, and I've updated this PR in the way you suggest. cc-ing @alimanfoo since the value 0.99999 came from scikit-allel's implementation: https://github.com/cggh/scikit-allel/blob/master/allel/stats/selection.py#L1323
There was a problem hiding this comment.
I've marked this as automerge as I doubt there was any deep reason for the 5-digit 0.9999.
Codecov Report
@@ Coverage Diff @@
## master #368 +/- ##
==========================================
- Coverage 95.53% 95.18% -0.35%
==========================================
Files 31 31
Lines 2239 2266 +27
==========================================
+ Hits 2139 2157 +18
- Misses 100 109 +9
Continue to review full report at Codecov.
|
Fixes #230