Carlos Parada

> I've updated the description of analytic weights in the light of #758. Does that sound correct? I think we should hold off until #758 is resolved.

Oh, brief note that I think could be useful for users -- currently, all our methods for `ProbabilityWeights` normalize the weights before calculating an estimator. This is probably the best...

I think it's an improvement, yeah, but I'd clarify that the weights: 1. Refer specifically to sample sizes for each observation. 2. I'd add a warning about `std` doing something...

> AFAIK all definitions for frequency weights generalize perfectly to non-integer weights. The only place where we check that weights are integers is in `quantile`, right? See #436. I think...

The `n-k-1` estimator is unbiased. The current implementation shrinks large lags to 0 by dividing by `n`, which usually reduces mean squared error. The major problem with the `n-k` estimator...

> Are you envisioning something like > > ```julia > std(x; correction=-1, mean=mean(x)) = sum(xi -> (xi - mean)^2, x) / (n + correction) > ``` > > ? Or...

> How would the stats functions typically work for weight matrices? Would it be comparable to broadcasting the weighted stats function across both weights and data (either colwise or rowwise)?...

> Could you show a concrete example where you would use weight matrices? The current use case is MCMC samples where I have a set of weights that can vary...

> Yes, if you write `@tullio grad=false` with no expression to calculate, this should update a global setting, which will apply to all subsequent macro calls. (It isn't per-module or...

@chriselrod Can you let me know when the fix is merged so I can close this?