[diminou]

Cool read, how do you convert something like logloss into something understandable by a ML contest noob? For instance it'd be cool to know the % of the time one of these algos predicted correctly, how do you get that from the logloss?

[richard_craib]

You actually can't get that from the logloss number directly. However, ln(0.5) = 0.693... That's the logloss you'll get if you guessed randomly ie had no edge.