Matthew Kay

Basic versions of this work (except shear). Will need to look into how the `justification` parameters work on viewports to allow scaling/rotation around a different point than the center of...

Thanks for raising this - somewhere on my todo list is working through a complete example comparing these (and other) different marginal effects, how to calculate them and what they...

Awesome!! I have a couple of thoughts: - If the symbolic derivative fails, can it fall back silently to numerical differentiation, instead of printing a message? Or at least, can...

Yeah, that pattern is used in a number of the transformations defined in {scales}. A simple example if you need a test case: ```r f = (function() { b =...

> that's perfect, thanks, it's good to have cases like this in the automatic tests anyway. You may be right, but let me see first if there is an easy...

Ah okay, I see why your solution works where you are applying it currently: within `Math` or `Ops` any constants / variables from the surrounding function will have already been...

In fact, this just made me realize that a simple way to solve the remaining issues with point 3 is, if finding the symbolic derivative of the substituted/composed inverse fails...

Maybe you could do something like store the functions as list within a single `dist_transformed()`, with `NULL` indicating "do numerical diff at this step". Or, if nesting the transformed distributions...

Just wanted to add my thanks @venpopov for getting this done! Very exciting

Ah yes, the vectorization of quantile/cdf/pdf was another thing I wanted to ask about! :) The data frame approach is neat, though I wonder how that generalizes down to replicating...