Joseph Egan, I believe

The one-vs-one AUC is implemented by R package called "HandTill2001"

https://cran.r-project.org/web/packages/HandTill2001/HandTill2001.pdf

Thank you Ljubomir!

That's ok :D I love multiple charts so multiple ROCs is ok for me :D

Do you have to have a probability for ROC? (Sorry for this a dumb question, but I been trying to figure out the proper way to do ROC when my predictor is ordinal)

You don't need probability, but you do need to be able to sort the cases by the score.

:O ? really? Then why does in always complain about needing a probability when I try SVM or (in the case of my question) ordinal logistic regression.

\o/ \o/

Thanks Ljubomir for the link to the "HandTill2001" R package for one-vs-one AUC!

could be due to software? it’s theoretically possible to construct ROC as long as you have ordered scores

logo is still a thing in minecraft

the computer in Minecraft do logo

The terms Mario mentioned ‘One-vs-Rest and ‘One-vs-One’ are also known as ‘One-Against-All’/’OAA’ and ‘One-Against-One’/’OAO’

@Kenneth let me watch and try to understand

Man this awesome.

`that turtle animation has really made this come alive..i am able to relate to this so much better after having seen that

+1

+1

+10

Interesting do you have the code for some these more advanced ROC curves? :D

You can also think of each orange box as a little ROC plot. The line is diagonal because in that subpopulation we have no information—all the scores are the same.

Brilliant!

+1!!

btw, the wilcoxon interpretation is the basis for the delong test, which is used to test the hypothesis that two AUCs are equal

OPTIMAL :O

python is evil! (jokes)

What is Zella?

zeallot

package for multiple assignment

@JohnMount: don’t understand why false_negative_value = 0 isn’t good enough?

Im not seeing the screen

?-?

I can see

working for me

Looks fine to me.

please continue

No sound from Nina.

Can some explain prevalence?

Prevalence threshold

prevalence is the proportion of the population that is positive