Switching linear dynamical systems #15

slinderman · 2021-11-24T00:44:57Z

TODO:

Implement Poisson emissions with stochastic M-step
Implement Laplace update for non-conjugate SLDS
Implement exact ELBO calculation for Gaussian SLDS (optional?)
Clean up the SLDS notebook and move expected log likelihood test at end of notebook to its own test file
Add a joint distribution over the initial discrete and continuous state p(z_1, x_1) = p(z_1) p(x_1 | z_1)
Clean up the SLDS property getters
Add SLDS tests

… ELBO to decrease transiently after the first iteration.

…ntation of the starfish dynamics when using the default pca-kmeans initialization.

github-actions · 2021-11-24T00:47:45Z

Unit Test Results

  1 files   1 suites 3m 1s ⏱️
34 tests 34 ✔️ 0 💤 0 ❌

Results for commit 282a4a4.

♻️ This comment has been updated with latest results.

- Adding mvn tests - Rewriting mvn log normalizer to remove time index from within scna - Switching PoissonGLM to use softplus link - Adding small positive bias to mvn message passing code

…iance

slinderman added 17 commits November 7, 2021 17:34

Starting an SLDS implementation.

a0a6784

Merge branch 'main' into scott.slds

8fb3cd9

scratching out some slds work

ef9b5c0

Fixing merge and making NIW and MNIW objects jittable.

25a1c3b

plumbing covariates and metadata.

74090dc

working on slds posterior. adding cavi updates

26d45cc

starting to work on slds init

25d6f30

Merge branch 'main' into scott.slds

851c568

adding emissions shape to PoissonEmissions

d3f7550

merging main

a041872

working on VI for GaussianSLDS

de4a97f

Variational EM for SLDS is running but theres still a bug causing the…

57941b9

… ELBO to decrease transiently after the first iteration.

adding missing files

d391926

Getting the SLDS to "work," though it fails to find the desired segme…

3389b27

…ntation of the starfish dynamics when using the default pca-kmeans initialization.

notebook cleanup

72dfe6e

cranking up the scale made a big difference!

c5194e8

Merge branch 'main' into scott.slds

abbf95d

slinderman added 9 commits November 24, 2021 08:23

fixing lds test bugs

6333f5c

moving expected_log_prob check to tests

b32ec86

cleaning up the SLDS notebook and variational EM callback support

2f5728b

adding laplace update for slds continuous posterior

ca38b57

basic poisson slds working

10c2f7a

updating poisson lds test since we switched the PoissonGLM link

cc69684

Working on Poisson SLDS

a7132e7

- Adding mvn tests - Rewriting mvn log normalizer to remove time index from within scna - Switching PoissonGLM to use softplus link - Adding small positive bias to mvn message passing code

reverting to the 10d Poisson SLDS example, which sorta works.

740f4ed

updating gaussian lds test since I changed the defaul constructor var…

282a4a4

…iance

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Switching linear dynamical systems #15

Switching linear dynamical systems #15

slinderman commented Nov 24, 2021 •

edited by andrewwarrington

Loading

github-actions bot commented Nov 24, 2021 •

edited

Loading

Switching linear dynamical systems #15

Are you sure you want to change the base?

Switching linear dynamical systems #15

Conversation

slinderman commented Nov 24, 2021 • edited by andrewwarrington Loading

github-actions bot commented Nov 24, 2021 • edited Loading

Unit Test Results

slinderman commented Nov 24, 2021 •

edited by andrewwarrington

Loading

github-actions bot commented Nov 24, 2021 •

edited

Loading