vectorize Jacobians in transforms

[alashworth]

Issue by bob-carpenter Thursday Jan 30, 2014 at 22:14 GMT Originally opened as https://github.com/stan-dev/stan/issues/524

---

Vectorize all the Jacobians. They should already be analytic, but if they're not, do that, too.

[alashworth]

Comment by syclik Wednesday Nov 30, 2016 at 15:16 GMT

---

@bob-carpenter, do we need to:

• add vectorized function signatures for Jacobians in the math library?
• add precomputed gradients for Jacobians in the math library?
• have code generator generate vectorized Jacobian calls? or do the two things above take care of it?

[alashworth]

Comment by bob-carpenter Wednesday Nov 30, 2016 at 20:04 GMT

---

@bob-carpenter, do we need to:

• add vectorized function signatures for Jacobians in the math library?

Yes. Vectorized versions of the transform function that take a single Jacobian to increment. It's the autodiff variable for that result that matters. E.g.,

real sigma[K];

We'd want to create a vari object to add to the log Jacobian, but we need to return K transformed variables.

• add precomputed gradients for Jacobians in the math library?

Yes. So this is going to need an issue there, too.

• have code generator generate vectorized Jacobian calls? or do the two things above take care of it?

Yes. That's mainly what needs to happen in this repo.