log regr with brier loss

[berndbischl]

we should say that this can also be naively considered

we have a nice exercise in optimml, which shows that this is not necessarly convex.

this has also sometimes in the old days discussed as an option for ANNs, and downsides have been discussed there

[ludwigbothmann]

bsp zur minimierung von den losses bei lt1 eibauen.
squared loss für classification noch einbauen?
squared loss auf f, oder pi(x)?
beides sinnvoll?

[tpielok]

relevant links

https://towardsdatascience.com/why-using-mean-squared-error-mse-cost-function-for-binary-classification-is-a-bad-idea-933089e90df7