Add ManifoldGaussian class for messages in belief propagation #121

joeaortiz · 2022-03-16T19:18:47Z

Motivation and Context

It would be useful to have an optional covariance / precision matrix as part of the Manifold class as Gaussian Belief Propagation involves sending Gaussian distributions over the Manifold variables. Currently I'm using a wrapper Gaussian class but it could be more widely useful to have covariance / precision matrix as an attribute of the Manifold class?

How Has This Been Tested

Types of changes

Docs change / refactoring / dependency upgrade
Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to change)

Checklist

My code follows the code style of this project.
My change requires a change to the documentation.
I have updated the documentation accordingly.
I have read the CONTRIBUTING document.
I have completed my CLA (see CONTRIBUTING)
I have added tests to cover my changes.
All new and existing tests passed.

…riance

theseus/geometry/manifold.py

joeaortiz · 2022-04-07T09:26:06Z

I assume that all Manifold instances passed in the mean Sequence have the same batch size, dtype and device.
I haven't included the get and set item functions as the class stores two internal properties: mean and precision.

theseus/optimizer/marginals.py

theseus/optimizer/manifold_gaussian.py

mhmukadam

Looks good so far! On to unit tests. Any other features you are considering adding still?

joeaortiz · 2022-04-08T15:07:50Z

Looks good so far! On to unit tests. Any other features you are considering adding still?

I don't think so. I was thinking of adding the local and retract functions but generalised to gaussians on manifold variables, but they would be better placed in the Manifold class.

mhmukadam · 2022-04-08T22:06:32Z

local and retract functions but generalised to gaussians on manifold variables

Okay probably add it in this PR. Manifold already has a local and retract, how will this Gaussian version get incorporated?

theseus/optimizer/manifold_gaussian.py

theseus/geometry/se3.py

theseus/optimizer/manifold_gaussian.py

theseus/optimizer/tests/test_manifold_gaussian.py

mhmukadam

Looks ready once we decide whether to move the local/retract inside the class.

theseus/optimizer/manifold_gaussian.py

mhmukadam

LGTM! Is the plan to keep local/retract outside of the class?

joeaortiz · 2022-04-19T07:38:22Z

Should be ready to merge now.

luisenp

Looks good! Left some comments/questions. I didn't check the math closely.

theseus/optimizer/manifold_gaussian.py

theseus/optimizer/tests/test_manifold_gaussian.py

theseus/optimizer/manifold_gaussian.py

mhmukadam

Look good to merge, nice work!

luisenp

LGTM!

* Change unit tests to avoid making mypy a main requirement. (#168) * Change unit tests to avoid making mypy a main requirement. * Added back dev requirements to CI. * Wording change. * Update readme and contrib (#169) * Change unit tests to avoid making mypy a main requirement. * Added back dev requirements to CI. * Removed thanks file. * Add missing license header in test_urdf_model * Add ManifoldGaussian class for messages in belief propagation (#121) * Gaussian class to wrap Manifold class and lam matrix for inverse covariance * reformatted * restored original manifold file * initial attempt at marginal class , need to handle batch dim * added standard fns dtype, to, copy, update * single to call in init * renamed ManifoldGaussian * setting precision in init with checks * update function requires mean and precision * fixed naming in init * manifold gaussian tests * retract and local gaussian fns * check precision is a symmetric matrix * moved retract and local gaussian to manifold_gaussian to avoid circular imports * added ManifoldGaussian to inits * minor edits * fixed dtype error in se3 that appeared in unit tests * add checks for local_gaussian * tests for local and retract gaussian * import from th. * added local_gaussian retract_gaussian to init, minor fix * minor fix on value error * fixed copy bug and added comments * random precision matrix in unit tests * fix for random precision * init precision with identity * fixed typo * More efficient implementation of forward kinematics (#175) * Bump DRM version and adapt API * Add small tolerences to tests to account for numerical errors * Changing setup virtualenv command. (#178) * Updated SDF object in collision cost functions whenever an aux var is updated (#177) * Updated SDF object in collision cost functions whenever an aux var is updated. * Changed gather_from_rows_cols to support matrix broadcasting. Co-authored-by: Mustafa Mukadam <[email protected]> Co-authored-by: Joe Ortiz <[email protected]> Co-authored-by: Austin Wang <[email protected]>

* Gaussian class to wrap Manifold class and lam matrix for inverse covariance * reformatted * restored original manifold file * initial attempt at marginal class , need to handle batch dim * added standard fns dtype, to, copy, update * single to call in init * renamed ManifoldGaussian * setting precision in init with checks * update function requires mean and precision * fixed naming in init * manifold gaussian tests * retract and local gaussian fns * check precision is a symmetric matrix * moved retract and local gaussian to manifold_gaussian to avoid circular imports * added ManifoldGaussian to inits * minor edits * fixed dtype error in se3 that appeared in unit tests * add checks for local_gaussian * tests for local and retract gaussian * import from th. * added local_gaussian retract_gaussian to init, minor fix * minor fix on value error * fixed copy bug and added comments * random precision matrix in unit tests * fix for random precision * init precision with identity * fixed typo

…okresearch#121) * Gaussian class to wrap Manifold class and lam matrix for inverse covariance * reformatted * restored original manifold file * initial attempt at marginal class , need to handle batch dim * added standard fns dtype, to, copy, update * single to call in init * renamed ManifoldGaussian * setting precision in init with checks * update function requires mean and precision * fixed naming in init * manifold gaussian tests * retract and local gaussian fns * check precision is a symmetric matrix * moved retract and local gaussian to manifold_gaussian to avoid circular imports * added ManifoldGaussian to inits * minor edits * fixed dtype error in se3 that appeared in unit tests * add checks for local_gaussian * tests for local and retract gaussian * import from th. * added local_gaussian retract_gaussian to init, minor fix * minor fix on value error * fixed copy bug and added comments * random precision matrix in unit tests * fix for random precision * init precision with identity * fixed typo

Gaussian class to wrap Manifold class and lam matrix for inverse cova…

78d991e

…riance

joeaortiz requested review from luisenp and mhmukadam March 16, 2022 19:18

facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Mar 16, 2022

reformatted

978e150

mhmukadam reviewed Mar 16, 2022

View reviewed changes

theseus/geometry/manifold.py Outdated Show resolved Hide resolved

mhmukadam assigned joeaortiz Mar 16, 2022

mhmukadam added the refactor Refactor library components label Mar 16, 2022

joeaortiz added 5 commits April 4, 2022 14:58

restored original manifold file

5e31531

Merge branch 'main' into joe.add_manifold_covariance

10389ec

initial attempt at marginal class , need to handle batch dim

312bd38

Merge branch 'main' into joe.add_manifold_covariance

90572c8

added standard fns dtype, to, copy, update

19c9d34

mhmukadam reviewed Apr 7, 2022

View reviewed changes

theseus/optimizer/marginals.py Outdated Show resolved Hide resolved

joeaortiz added 2 commits April 7, 2022 15:34

single to call in init

167bb84

renamed ManifoldGaussian

572486a

mhmukadam reviewed Apr 7, 2022

View reviewed changes

theseus/optimizer/marginals.py Outdated Show resolved Hide resolved

mhmukadam reviewed Apr 7, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam marked this pull request as ready for review April 7, 2022 23:23

setting precision in init with checks

4cc5f78

mhmukadam reviewed Apr 8, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam reviewed Apr 8, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam reviewed Apr 8, 2022

View reviewed changes

update function requires mean and precision

a8ecb77

mhmukadam reviewed Apr 8, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

fixed naming in init

23cc2f9

mhmukadam reviewed Apr 11, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Show resolved Hide resolved

mhmukadam reviewed Apr 13, 2022

View reviewed changes

theseus/geometry/se3.py Outdated Show resolved Hide resolved

import from th.

eecf1a7

mhmukadam reviewed Apr 13, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam reviewed Apr 13, 2022

View reviewed changes

theseus/optimizer/tests/test_manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam reviewed Apr 13, 2022

View reviewed changes

mhmukadam mentioned this pull request Apr 13, 2022

local and retract to route based on Manifold or ManifoldGaussian #159

Open

added local_gaussian retract_gaussian to init, minor fix

6812d9b

mhmukadam reviewed Apr 13, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam approved these changes Apr 13, 2022

View reviewed changes

minor fix on value error

8a97d10

mhmukadam changed the title ~~Add optional covariance / precision matrix to Manifold class~~ Add ManifoldGaussian class for messages in belief propagation Apr 14, 2022

mhmukadam added enhancement New feature or request and removed refactor Refactor library components labels Apr 14, 2022

Merge branch main into joe.add_manifold_covariance

218d754

luisenp requested changes Apr 19, 2022

View reviewed changes

joeaortiz added 4 commits April 19, 2022 17:18

fixed copy bug and added comments

8ac7967

random precision matrix in unit tests

e19a8d7

fix for random precision

4e98aff

init precision with identity

7528e60

mhmukadam reviewed Apr 19, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Outdated Show resolved Hide resolved

mhmukadam reviewed Apr 19, 2022

View reviewed changes

theseus/optimizer/manifold_gaussian.py Show resolved Hide resolved

fixed typo

6a7b1a4

mhmukadam approved these changes Apr 20, 2022

View reviewed changes

luisenp approved these changes Apr 20, 2022

View reviewed changes

mhmukadam merged commit 2769876 into main Apr 20, 2022

mhmukadam deleted the joe.add_manifold_covariance branch April 20, 2022 15:17

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add ManifoldGaussian class for messages in belief propagation #121

Add ManifoldGaussian class for messages in belief propagation #121

joeaortiz commented Mar 16, 2022

joeaortiz commented Apr 7, 2022

mhmukadam left a comment

joeaortiz commented Apr 8, 2022

mhmukadam commented Apr 8, 2022 •

edited

Loading

mhmukadam left a comment

mhmukadam left a comment •

edited

Loading

joeaortiz commented Apr 19, 2022

luisenp left a comment

mhmukadam left a comment

luisenp left a comment

Add ManifoldGaussian class for messages in belief propagation #121

Add ManifoldGaussian class for messages in belief propagation #121

Conversation

joeaortiz commented Mar 16, 2022

Motivation and Context

How Has This Been Tested

Types of changes

Checklist

joeaortiz commented Apr 7, 2022

mhmukadam left a comment

Choose a reason for hiding this comment

joeaortiz commented Apr 8, 2022

mhmukadam commented Apr 8, 2022 • edited Loading

mhmukadam left a comment

Choose a reason for hiding this comment

mhmukadam left a comment • edited Loading

Choose a reason for hiding this comment

joeaortiz commented Apr 19, 2022

luisenp left a comment

Choose a reason for hiding this comment

mhmukadam left a comment

Choose a reason for hiding this comment

luisenp left a comment

Choose a reason for hiding this comment

mhmukadam commented Apr 8, 2022 •

edited

Loading

mhmukadam left a comment •

edited

Loading