Find where second-order is failing and create further issues to fix

[bob-carpenter]

Summary:

@syclik claims that if we try to build all the models in stan-dev/stan/test-models/good for fvar<var> that they will break because second-order autodiff isn't complete.

Description:

These are the functions that we can't compile using stan::math::fvar<double> and/or stan::math::fvar<stan::math::var>:

• stan/math/prim/scal/fun/inc_beta.hpp
• stan/math/prim/mat/fun/quad_form.hpp
• prim/mat/fun/trace_gen_quad_form.hpp

And all the ODE functionality:

• stan/math/rev/mat/functor/cvodes_ode_data.hpp
• stan/math/rev/arr/fun/decouple_ode_states.hpp
• stan/math/rev/mat/functor/integrate_ode_bdf.hpp
• stan/math/rev/mat/functor/cvodes_ode_data.hpp
• stan/math/rev/arr/fun/decouple_ode_states.hpp
• stan/math/prim/arr/functor/integrate_ode_rk45.hpp

Reproducible Steps:

Need to hack the test in stan-dev/stan to compile for second-order or do something else similar like create a new make target.

Current Output:

Failing?

Expected Output:

Everything buildable.

Current Version:

v2.12.0

[syclik]

It is failing. I need to figure out how to run through all the tests, but the first to fail: asinh can't be instantiated with fvar<fvar<var>>.

[bgoodri]

Should be able to pass the -k flag to make inside runTests.py.

[syclik]

I'm editing the first post to have a checklist of the failing instantiations.

[syclik]

For getting us to compile the Stan language with second-order autodiff, we're not as bad as I thought. We're missing some functions and the ODEs. The ODEs might take some effort to untangle. Since the introduction of CVODES, the code has gotten less general and more dependent on stan::math::var.

[bob-carpenter]

I think we should figure out a way to leave out ODEs to start if we can.

[syclik]

Agreed. That's why I split out the list. Anyone with time could tackle the first 3 functions.

…

On Tue, Nov 29, 2016 at 9:21 PM, Bob Carpenter ***@***.***> wrote: I think we should figure out a way to leave out ODEs to start if we can. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#403 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAZ_F1uv-1CDJCVB2GPaEhJdM9z7C2xzks5rDN2PgaJpZM4KN98B> .