How to suppress the warning message for binary classification metrics?

[liuyxpp]

I am using MLJ v0.20.2 with StatisticalMeasures.jl v0.1.3.

yt = Int.(rand(100) .> 0.5)
yp = Int.(rand(100) .> 0.5)
cm = MLJ.confmat(yp, yt)
MLJ.PositivePredictiveValue()(cm)

which emitts a warning:

┌ Warning: Levels not explicitly ordered. Using the order [0, 1]. The "positive" level is 1. 
└ @ StatisticalMeasures.ConfusionMatrices [~/.julia/packages/StatisticalMeasures/NoDLI/src/confusion_matrices.jl:333](https://vscode-remote+ssh-002dremote-002bivosrv6.vscode-resource.vscode-cdn.net/home/lyx/projects/rules/~/.julia/packages/StatisticalMeasures/NoDLI/src/confusion_matrices.jl:333)

Many other measures have similar behavior. I try to add warn=false keyword (which was found in the above mentioned confusion_matrices.jl but without success.

Is there any easy way to get rid of this warning. Or better, how can I correctly set the levels to eliminate the warning? I try a lot of ways but failed. Thanks!

[ablaom]

1. Thanks for your question.

using StatisticalMeasures

# option 1: Use categorical arrays for categorical data (recommended throughout MLJ ecosystem)
using CategoricalArrays
yt = categorical(yt, ordered=true) # ordered 0, 1
yp = categorical(yp, ordered=true)
cm = confmat(yp, yt)
PositivePredictiveValue()(cm)
# 0.5306122448979591

# option 2: Notice `confmat` is just alias for `ConfusionMatrix()`, which has `levels` as
# an option; see its docstring.
yt = Int.(rand(100) .> 0.5)
yp = Int.(rand(100) .> 0.5)
confmat2 = ConfusionMatrix(levels=[0, 1]) # negative class first
cm = confmat2(yp, yt)
PositivePredictiveValue()(cm)
# 0.5306122448979591

Is this a satisfactory answer?

[ablaom]

And of course, the choice of negative class matters:

confmat2 = ConfusionMatrix(levels=[1, 0]) # negative class first
cm = confmat2(yp, yt)
PositivePredictiveValue()(cm)
# 0.5087719298245614

[liuyxpp]

Crystal & clear, Thanks! I will follow the recommended option 1.