MINT output BER vs overall is swapped?

Create issue
Issue #141 resolved
Kim-Anh Le Cao repo owner created an issue

It seems to me that the numerical outputs from MINT should be swapped (BER/overall), see the stemcells case study (code are on bitbucket in Wordpress/) and screenshot attached

Could you double check, and also check for the other methods output? Thanks Florian!! Screen Shot 2018-05-18 at 10.43.10 am.png

Comments (5)

  1. Florian Rohart

    BER max.dist is in solid purple line: comp1 0.51, comp2 0.37

    overall max.dist in is dotted purple line: comp1 0.56, comp2 0.41

    dotted purple line higher than solid purple line, which is what we see in the graph. no?

  2. Kim-Anh Le Cao reporter

    I was thinking more about the output in general. I was surprised to see that the BER < overall error rate. I had a user recalculating the weighted error rates and maybe the numbers have been swapped (i.e. BER should be 0.568 on comp 1 max dist instead of 0.51289).

  3. Florian Rohart

    Well, I don't know how that user calculated BER and overall errors, but the code checks out:

                conf = get.confusion_matrix(truth = factor(Y), predicted = x)
                BER = get.BER(conf)
                overall = sum(apply(conf, 1, sum) - diag(conf)) / length(Y)
  4. Log in to comment