[WIP] Embedded Laplace Approximation #3097
Conversation
…4.1 (tags/RELEASE_600/final)
…4.1 (tags/RELEASE_600/final)
… into experimental/laplace
| struct laplace_options { | ||
| /* Size of the blocks in block diagonal hessian*/ | ||
| int hessian_block_size{1}; | ||
| /** | ||
| * Which Newton solver to use: | ||
| * (1) method using the root of W | ||
| * (2) method using the root of the covariance | ||
| * (3) method using an LU decomposition | ||
| */ | ||
| int solver{1}; | ||
| /* Maximum number of steps in line search*/ | ||
| int max_steps_line_search{0}; | ||
| /* iterations end when difference in objective function is less than tolerance | ||
| */ | ||
| double tolerance{1e-12}; | ||
| /* Maximum number of steps*/ | ||
| int64_t max_num_steps{100}; | ||
| }; |
There was a problem hiding this comment.
@charlesm93 can you check these defaults are correct?
|
|
|
Before I cut out a lot of tests I want to try building a separate jenkins node that just compiles these. At O3 these tests run decently quickly |
Jenkins Console Log Machine informationNo LSB modules are available. Distributor ID: Ubuntu Description: Ubuntu 20.04.3 LTS Release: 20.04 Codename: focalCPU: G++: Clang: |
Summary
Code for the embedded laplace approximation. The tests are all passing but there is a few things still needed
The file for laplace have been added to the
mixfolder since it uses higher order auto diff.The current signature for the generalized laplace looks like the following in C++
which will translate to stan like the following
Note that the first tuple used for the likelihood arguments must be data.
Instead of using a tuple for the first functor's inputs and variadic arguments for the covariance functors arguments I would rather have them both be tuples like the following
I think this is nice because it makes it makes the tolerance parameters always sit at the end and both functors have the same input scheme for their arguments. Does anyone have thoughts on this
Other additions related to this PR
filter_mapfunction that applies a conditionally applies a lambdafto each input of a tuple given atype_traiti.e. the following code would print "fp detected" twice and increment thedoubleelements of the tuple by 1.The test_ad suite now has a compile time option for only running the tests with only prim and reverse mode with a new boolean template parameter to
expect_ad. This is needed to use laplace with the test framework as the laplace impl here does not work with higer order autodiff (since it needs higher order autodiff)Tests
Since the tests all seem very related I kept them in their own folder, is that alright? Or should I distribute them across the test folders like normal? While this PR is WIP I'm going to leave them in the same folder and if we don't want that then we can move them before we merge
Side Effects
Release notes
Checklist
Copyright holder: Simon's Foundation
The copyright holder is typically you or your assignee, such as a university or company. By submitting this pull request, the copyright holder is agreeing to the license the submitted work under the following licenses:
- Code: BSD 3-clause (https://opensource.org/licenses/BSD-3-Clause)
- Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
the basic tests are passing
./runTests.py test/unit)make test-headers)make test-math-dependencies)make doxygen)make cpplint)the code is written in idiomatic C++ and changes are documented in the doxygen
the new changes are tested