Ryan
The nested list structure returned by `GetInferenceSignals` and `RerunForSignals` is unnecessarily complicated. Replace it with tibbles with a list-valued column, e.g.: ``` these_signals
The new `GetInferenceSignals` sometimes throws the following harmless error due to our use of `approx` with possibly very small influence scores: ```Error in regularize.values(x, y, ties, missing(ties)) : (converted from...
I think `_order1` and `_order2` are bigger than they need to be. Perhaps I was careless when setting it up for forward mode because there was no cost.
We can use a reverse mode Taylor series expansion instead.