Need snippet for confusion matrix

1

How does one use tnt.meter.ConfusionMeter? It would be good if a snippet is available.

2

just used it a few minutes ago in a classifier network

    confusion_matrix = meter.ConfusionMeter(2) #I have 2 classes here
    for ii, data in enumerate(val_dataloader):
        input, label = data
        val_input = Variable(input, volatile=True).cuda()
        val_label = Variable(label.type(t.LongTensor), volatile=True).cuda()
        score = model(val_input)
        confusion_matrix.add(score.data.squeeze(), label.type(t.LongTensor))
    print confusion_matrix.conf

got something like:

[[1152, 434],
 [764, 722]]

that is

[true negative, false postive]
[false negative,ture postive]
6 Likes
3

Thank you. I figured that I had to convert the Autograd Variable back into a tensor for ConfusionMeter to do the .numpy() conversion to get it to work properly.

4

score and label are Autograd variables so
Confusion_matrix.add(score.data, label.data) will work