the order of classification

[wangjiawen2013]

Hi,
here is a confusion matrix getting from logistic regression:

classes	negative	positive
negative	295	0
positive	20	4

But I expect it is:

classes	positive	negative
positive	0	295
negative	4	20

How does linfa determine the order of the classification ? Does it affect the calculation of accuracy and decision ? which column is ground truth and which column is prediction ?

[wangjiawen2013]

Besides, I find that the confusion_matrix is not stable, it varies each run ! Is it caused by HashSet whose element is not in fixed order ?

pub trait Labels {
    type Elem: Label;

    fn label_count(&self) -> Vec<HashMap<Self::Elem, usize>>;

    fn label_set(&self) -> Vec<HashSet<Self::Elem>> {
        self.label_count()
            .iter()
            .map(|x| x.keys().cloned().collect::<HashSet<_>>())
            .collect()
    }

Run the program first time:

Run the program again: