@MattBMcDermott 3. Do you have any thoughts or opinions on Hand's criticisms of AUROC? https://t.co/YXmojfDVXq https://t.co/DF5lNa0KDr
@therriaultphd @rasbt @tunguz @svpino @JFPuget Ah you’re referring to whether the AUC is a rank statistic? If so it is indeed a Mann-Whitney-Wilcoxon U statistic. The probabilistic statement I gave is one way of defining the AUC so not sure I follow? Wh
@dan_p_simpson Hand's 2009 paper isn't last-couple-of-years recent, but maybe it's be useful? https://t.co/ZcRzuavd6G
@GaelVaroquaux @martin_hebart @danilobzdok @bttyeo @agramfort @KordingLab @tyrell_turing @dngman @ten_photos agree with @GaelVaroquaux here too, though i would point out that the ROC curve is also problematic: https://t.co/PD7d5Td191 in general, i like Co
RT @NeuroStats: Then there are the sorts of issues pointed out by Hand. "Classifier Technology and the Illusion of Progress" https://t.co/…
Then there are the sorts of issues pointed out by Hand. "Classifier Technology and the Illusion of Progress" https://t.co/4o6xIiIs0D The AUC is "fundamentally incoherent in terms of misclassification costs" https://t.co/tJDvnuz37M
RT @RogueRad: ROC's classification problem. Is there an alternative? https://t.co/jtZfN9Wkh9 (arresting line: " It is as if one chose to co…
Measuring classifier performance: a coherent alternative to the area under the ROC curve https://t.co/9S2PJCssp0
RT @RogueRad: ROC's classification problem. Is there an alternative? https://t.co/jtZfN9Wkh9 (arresting line: " It is as if one chose to co…
ROC's classification problem. Is there an alternative? https://t.co/jtZfN9Wkh9 (arresting line: " It is as if one chose to compare the heights of two people using rulers in which the basic units of measurement themselves depended on the heights")
Not an issue on @kaggle since everyone is using #xgboost anyways https://t.co/VKWlLgnBE5
I'm reading this on #springerlink https://t.co/wkSYdQzjiY
@deaneckles @johnmyleswhite Also overall challenges with ROCs because people don't distinguish population vs. sample https://t.co/xmvApQcETO
.@stephensenn We only get to work with sample ROCs. Drawbacks usually unacknowledged in classic ML E.g. https://t.co/muVIobFNBq https://t.co/XmXIEqfL95
RT @neilfws: Enjoyed this and was reminded of this paper on deficiencies of ROC AUC https://t.co/7DZg0qCqDJ https://t.co/cm7YEuIKwN
RT @neilfws: Enjoyed this and was reminded of this paper on deficiencies of ROC AUC https://t.co/7DZg0qCqDJ https://t.co/cm7YEuIKwN
Enjoyed this and was reminded of this paper on deficiencies of ROC AUC https://t.co/7DZg0qCqDJ https://t.co/cm7YEuIKwN
RT @ScienceGuyRob: @kdnuggets not always, this paper by David J. Hand describes some of its deficiences https://t.co/f4djwd5q4K
RT @ScienceGuyRob: @kdnuggets not always, this paper by David J. Hand describes some of its deficiences https://t.co/f4djwd5q4K
RT @ScienceGuyRob: @kdnuggets not always, this paper by David J. Hand describes some of its deficiences https://t.co/f4djwd5q4K
RT @ScienceGuyRob: @kdnuggets not always, this paper by David J. Hand describes some of its deficiences https://t.co/f4djwd5q4K
RT @ScienceGuyRob: @kdnuggets not always, this paper by David J. Hand describes some of its deficiences https://t.co/f4djwd5q4K
@kdnuggets not always, this paper by David J. Hand describes some of its deficiences https://t.co/f4djwd5q4K
Not always. I recall a paper by David J. Hand (2009) that's describes its deficiencies: https://t.co/8l53BhjRPi https://t.co/GjISqdSPcc
Measuring classifier performance, AUC isn't good either http://t.co/aDR92UcQ