pgree

@bgoodri @wds15 @bbbales2 and I were hoping to get your thoughts on an interpolation algorithm for which I just made a pull request here . @bbbales2 mentioned that the need...

Thanks for all the comments. We decided to implement a scheme in which the interpolated function is differentiable based on several comments/requests on discourse and github: https://discourse.mc-stan.org/t/linear-interpolation-and-searchsorted-in-stan/13318/1 https://discourse.mc-stan.org/t/vectors-arrays-interpolation-other-things/10253/17 https://github.com/stan-dev/stan/issues/1165 https://github.com/stan-dev/stan/issues/1214...

> We should make everything differentiable if it's just an ordinary Stan program. I will look into this. Right now, the codes only provide derivatives of f(t) with respect to...

> Did you mean smoothness or regularization? By smoothness I meant the size of the standard deviation in the Gaussian kernel in the convolution. > I know how to interpret...

I've updated the Guide-Level Explanation section of the design doc [here](https://github.com/pgree/design-docs/blob/feature/interp/designs/0018-interp.md) with a couple of Stan code snippets showing possible implementations of linear interpolation and smooth interpolation. @andrewgelman for the...

@nhuurre - thanks for the numerical experiments and the interpolation scheme. One environment in which the Gaussian-smoothed interpolation can come in handy is when you have a smooth function tabulated...

> I'd assume either (-pi, pi) or (mu - pi, mu + pi) would have been the two obvious choices. > The question is then whether we take mu modulo...

I think a lot of these choices depend heavily on how the von Mises distribution would be used by Stan users. You all have a much better sense of this...