(ML 11.3) Frequentist risk, Bayesian expected loss, and Bayes risk

A simple way to visualize the relationships between the frequentist risk, Bayesian expected loss, and Bayes risk.

[@marvinpolo6106]

him: "what makes it confusing is a bunch of bad notation" *starts explaining by introducing bunch of notation" ...interesting

[@sairushikjasti660]

"what makes it confusing is a bunch of bad notations" me watching this video and not understanding a single notation, welp

[@frobeniusfg]

This is your last chance. After this there is no turning back. You take the red pill: the story ends, you wake up in your bed and believe whatever frequentists want you to believe. You take the blue pill: you stay in Bayesland and I show you how deep the rabbit hole goes.

[@bztube888]

Great, intuitive explanation. Books like to attack with formulas before bother to try to explain the ideas behind them.

[@mojomagoojohnson]

Great!

[@auggiewilliams3565]

poor explanation

[@monsume123]

Great video! Love the symmetrical drawing and perfect coloring! I was wondering whether conceptually it would be more complete to include the Data D into the frequentist notation since it is also conditioning on the data in the frequentist perspective. In my understanding the frequentist perspective has both the parameter and the data as deterministic elements, where the parameter in the end is of course the optimisation variable but since you wrote that theta is a conditional dependence, so should be the data or am I conceptually confusing something here?

[@jiatianxu757]

Actually, we usually care about L(y-hat, y) which is L(f(theta hat), y)

[@jinghuayan5128]

what do you mean by average over theta? I am a little clueless about this. Looking forward to hear from you, thank you!

[@yugiohgx114]

Average over theta means that we multiply the loss function L(theta, f(D)) by the posterior probabilities p(theta | D) for all values of theta

[@danielmburu6936]

hello how can i get the posterior distribution of the poisson given the prior as e raised to power labda

[@jiatianxu757]

P(theta|D) is proportional for P(D|theta) * P(theta), in your case, it will be the Poisson mass function given theta * the prior dist.