tnr_ using same cutoff as confusionMatrix_

[labrax]

Hi Tyler,

Hope you are doing well. Thanks for this great package.

On https://github.com/JackStat/ModelMetrics/blob/master/src/confusionMatrix_.cpp:

• confusionMatrix_ considers predictions as negative if the value is <= 0
• tnr_ considers predictions as negative if the value is < 0.

I suggest to standardise it using confusionMatrix_, as in the attached patch.

I am not fully aware, but I strongly suggest checking the functions mcc_ and kappa_ as they might have a different result due to the different way it is calculated.

Best wishes,
Victor

[coveralls]

Coverage decreased (-0.1%) to 56.429% when pulling 8230b81 on labrax:master into d3cbb5f on JackStat:master.