Use std of 0 for singleton vectors #90
Conversation
Codecov Report
@@ Coverage Diff @@
## main #90 +/- ##
=======================================
Coverage 99.25% 99.25%
=======================================
Files 12 12
Lines 135 135
=======================================
Hits 134 134
Misses 1 1
Continue to review full report at Codecov.
|
src/scaling.jl
Outdated
| @@ -68,7 +68,8 @@ struct MeanStdScaling <: AbstractScaling | |||
| end | |||
| end | |||
|
|
|||
| compute_stats(x) = (mean(x), std(x)) | |||
| # Set std to 0 using corrected=false if x is a singleton | |||
| compute_stats(x) = (mean(x), std(x; corrected=(length(x) != 1))) | |||
There was a problem hiding this comment.
maybe it's better to pass this via MeanStdScaling as a kwarg? since it's already exposed via Statistics.std anyway?
There was a problem hiding this comment.
I see where you're coming from by leaving it up to the user. I guess it would just be inconvenient in the case that came up for me. I would rather change my downstream test case so that the data doesn't end up being one row, than add corrected=(length(x) != 1) to MeanStdScaling in the transform pipeline, in the source code.
There was a problem hiding this comment.
yeah I think it's better to not make those kinds of assumptions in the code here, especially when Statistics.std doesn't. In the instance of edge-case we just have to handle it on that end.
| @@ -31,8 +31,8 @@ struct MeanStdScaling <: AbstractScaling | |||
| σ::Real | |||
|
|
|||
| """ | |||
| MeanStdScaling(A::AbstractArray; dims=:, inds=:) -> MeanStdScaling | |||
| MeanStdScaling(table, [cols]) -> MeanStdScaling | |||
| MeanStdScaling(A::AbstractArray; dims=:, inds=:, corrected=true) -> MeanStdScaling | |||
There was a problem hiding this comment.
Can we just pass generic kwargs...? Or does it get confused about what to do with the dims kwarg?
Closes #89 by adding an optional kwarg
correctedwhich is passed to thestdfunction.