Bug with nanmean

[ph-kev]

I am using NaNStatistics v0.6.52 and encountered a bug with nanmean. See the minimal working example below.

using NaNStatistics
using Test

# Expected
small_data = fill(NaN, 2, 3, 4)
nan_mean_small_data = nanmean(small_data, dims = 2)
@test isequal(nan_mean_small_data, fill(NaN, 2, 1, 4)) # pass

# Not expected
large_data = fill(NaN, 180, 10, 360)
nan_mean_large_data = nanmean(large_data, dims = 2)
@test isequal(nan_mean_large_data, fill(NaN, 180, 1, 360)) # fail

The values in nan_mean_large_data are all zeros instead of NaNs.

[brenhinkeller]

Ah, thanks for reporting! @JamesWrigley any idea if this is related to #57

[JamesWrigley]

Yes, I wasn't thinking clearly 🤦

NaNStatistics.jl/src/ArrayStats/nanmean.jl

Line 130 in 7f92233

B[i] = ∅

Will make a PR to fix it.

[brenhinkeller]

Ok, fix is tagged -- v0.6.53