ml-explore
Change the variance computation
Changes the computation of variance to a faster one (but slightly less numerically stable). Not sure if we want to merge this but the performance benefit for large reductions is quite big.
What do you think about waiting for multi-output primitives and doing a primitive for this instead?
I think it may be worth doing a primitive but no reason for a multi output one imho.
Computing the mean and passing it as an argument to the variance primitive makes a lot of sense I think.
We can close it but what do we intend to do? Shall we do a primitive? And if yes why? Are we thinking of not merging due to possible numerical instability?