Detecting Anomalies Using Statistical Distances | SciPy 2018 | Charles Masson

[@arshamafsardeir2692]

The best explanation of Statistical Distances that I have found. Easy and nice explanation of Kolmogrov-Smirinov, Wasserstein distance, and KL-divergence.

[@mathman2170]

Love it when a talk presents the material in a carefully developed, logical, manner. Merci!

[@kimmupfumira3417]

Great explanation! Easy to digest.

[@jamesmckeown4743]

17:13 there should be a negative in the definition of KL

[@Mayur7Garg]

I think the negative should be based on whether you are minimizing or maximizing it. By definition, distances are always positive.

[@MauricioSalazare]

Well done! Nice explanation!

[@nmertsch8725]

This is a great presentation! Is there a reason why you did not commit the nth Wasserstein distance to SciPy?

[@canmetan670]

docs.scipy.org/doc/scipy/reference/generated/scipy.stats.wasserstein_distance.html As of this date, latest stable version of scipy is 1.3.1 on pip. This has been allegedly available after 1.0.0

[@TheBjjninja]

6:15 we should either reject or fail to reject H0 i believe. Instead of “accept H0”

[@harry8175ritchie]

AKA accept. I think it depends on where you learn statistics. My professors always said accept and reject.

[@mikhaeldito]

Semantically, "accepting H0" and "failing to reject H0" are the same. But they are not! P-value is a measure of the probability of our data assuming that the null hypothesis (such as no difference between two groups) is true. So, it is a measure against the null, not in favour of the null. This is why we have a statistical test of no difference, or similarity, that is called equivalence tests.

[@joelwillis2043]

@@harry8175ritchie AKA NO. You can't conclude your assumption based on your assumption. This is like logic 101. HARD FAIL GO DRIVE A TRUCK FOR LIVING.

[@harry8175ritchie]

Not the way to handle it buddy.

[@minesinitiativesrussie1778]

T'es le meilleur fillot ! J'ai rien compris mais c'est quand même la classe !

[@nomcognom2332]

Good!

[@112ffhgffg12]

Thanks

[@jesuse4691]

G