Hugues Fontenelle

Oslo, Norway

Age: 36

Mar
27
comment how to calculate roc curves?
The initial definitions are wrong. Precision != Sensitivity, please review.
Mar
25
awarded Scholar
Mar
25
awarded Benefactor
Mar
25
accepted Correcting naïve Sensitivity and Specificity for classifier tested against imperfect gold standard
Mar
24
comment Correcting naïve Sensitivity and Specificity for classifier tested against imperfect gold standard
Thank you for your answers. I was trying to avoid domain specifics, but since you asked, here are my beliefs: the data is related to genetics. The True Positive set comes from a database HGMD while the True Negative set is based on an assumption (which is, MAF>=0.20 is a benign variant). I have heard in conferences that the HGMD contained "errors" (but I cannot find citations nor proper quantification); while the benign assumption does not hold at all times.
Mar
19
revised Correcting naïve Sensitivity and Specificity for classifier tested against imperfect gold standard
title
Mar
19
awarded Promoter
Mar
17
comment Understanding Model Credibility with True/False Positive/Negative
I should have seen that. The problem is asymetric, in that many people do not buy the bike (the condition negative is prevalent). That's okay but the AUC tells you, correctly, that it is > 0.50, which is unhelpful. You're correct (but you may need to work on the model/features)
Mar
17
comment Understanding Model Credibility with True/False Positive/Negative
I would plot the ROC (receiver operator curve) of the classifier and see if it does better than random (the diagonal of that plot). If it does, then you're onto something. But the numbers can vary a lot between application. Sometimes it's better not to know anything than having predictions polluted by false positive (of lack of answers as in false negatives), sometimes any bit can help. I do medical testing and.. it depends :-)
Mar
17
comment Understanding Model Credibility with True/False Positive/Negative
Don't forget to up-vote and accept, I'm paid in points ;-)
Mar
17
comment Understanding Model Credibility with True/False Positive/Negative
Well, if there isn't any threshold that would give you much higher sensitivity, then I would argue that the problem lies with the feature vector, i.e. the data that you put in.
Mar
17
awarded Editor
Mar
17
revised Understanding Model Credibility with True/False Positive/Negative
edit 1: threshold
Mar
17
answered Understanding Model Credibility with True/False Positive/Negative
Mar
17
awarded Student
Mar
17
asked Correcting naïve Sensitivity and Specificity for classifier tested against imperfect gold standard
Mar
11
comment Scrapy - Scrape multiple URLs using results from the first URL
Is there anything wrong if you do?
Mar
10
awarded Supporter
Mar
6
comment Subcommand alternative to argparse and optparse
Oh now I see what you mean. Good idea, I'll start using those too.. once you get a good answer to your question :-)
Mar
6
comment Subcommand alternative to argparse and optparse
What it says on the tin is totally unhelpful. What it does is much better though.
1 2 3 4 5