Andreas Noack

Below is a more complete list and there are a couple of other places things could go wrong. I guess `x` and `y` might be non-negative in this application so...

Yes. That would leave some room when starting again after a `tstop`. However, isn't there still a risk that an artificially small step before a tstop makes the first step...

I'd prefer the second solution. We should also look into making the function thread safe.

Using the Cholesky is typically much faster and most often the statistical error is much larger than error due to roundoff so in many cases it can preferable with the...

Well, in your example you chose zero statistical error so, of course, the errors due to rounding will dominate. If you data matrix is practically singular and your explained variable...

> Are you saying that because estimation has statistical error anyway, we might as well ignore potentially significant numerical error? I think the fear of rounding errors of least squares...

Showing the values would be helpful. It would great if you could open a PR with your change.

Why don't we use call overloading of each distance type here?

Just to clarify my proposal. I think, it would be better with ``` julia> Euclidean()(randn(10), randn(10)) 4.28348729614181 ``` Are there any places where this would cause trouble?

I might try to take a look at it and maybe I can figure out what is going on. If not it might be better to disable.