Switching to Bridgestan and JAX #58

CyrusZhang73 · 2023-10-17T19:12:10Z

No description provided.

codecov-commenter · 2023-10-18T19:05:31Z

Codecov Report

Attention: 139 lines in your changes are missing coverage. Please review.

Comparison is base (7c6e5b1) 92.73% compared to head (8248b6f) 87.20%.

Files	Patch %	Lines
viabel/patterns.py	89.92%	84 Missing ⚠️
viabel/optimization.py	55.55%	32 Missing ⚠️
viabel/function_patterns.py	92.10%	9 Missing ⚠️
viabel/objectives.py	87.30%	8 Missing ⚠️
viabel/models.py	70.00%	6 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #58      +/-   ##
==========================================
- Coverage   92.73%   87.20%   -5.53%     
==========================================
  Files           6        8       +2     
  Lines        1211     2228    +1017     
==========================================
+ Hits         1123     1943     +820     
- Misses         88      285     +197

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

jhuggins · 2023-11-07T21:28:22Z

viabel/convenience.py



-def bbvi(dimension, *, n_iters=10000, num_mc_samples=10, log_density=None,
+def bbvi(dimension, *, n_iters=3000, num_mc_samples=10, log_density=None,


don't change defaults; if you need it to be different for the test, change it manually in the test

jhuggins · 2023-11-07T21:28:38Z

viabel/convenience.py



-def vi_diagnostics(var_param, *, objective=None, model=None, approx=None, n_samples=100000):
+def vi_diagnostics(var_param, *, objective=None, model=None, approx=None, n_samples=3000):


same here; don't change defaults

jhuggins · 2023-11-07T21:31:29Z

viabel/_mc_diagnostics.py

    rho_hat_t = np.zeros(n_draw)
    rho_hat_even = 1.0
-    rho_hat_t[0] = rho_hat_even
+    rho_hat_t =rho_hat_t.at[0].set(rho_hat_even)


add space after the equals sign

jhuggins · 2023-11-07T21:31:50Z

viabel/approximations.py

    return ms_pattern

-
+def sqrtm(matrix):


use cholesky instead

jhuggins · 2023-11-07T21:32:31Z

viabel/function_patterns.py

@@ -0,0 +1,393 @@
+import copy


credit paragami

jhuggins · 2023-11-21T17:47:58Z

viabel/optimization.py

        # if descent_dir_history is not None:
        #     results['descent_dir_history'] = descent_dir_history
-        results_dict = {d: np.array(h) for d, h in results.items()}
+        results_dict = {d: jnp.array(h) for d, h in results.items()}


why not use numpy here and below?

jhuggins · 2023-11-21T17:48:51Z

viabel/optimization.py

-        fit = model.sampling(data=data, init=init, iter=1000, chains=n_chains,
-                             control=dict(adapt_delta=0.98)) #sampling from the model
+        model = stan.build(program_code=model_code, data=data)
+        fit = model.sample(num_chains=n_chains, num_samples=1000,init = init)   # sampling from the model


is this really a fit? Or should it be called something else?

jhuggins · 2023-11-21T17:49:45Z

viabel/tests/test_convenience.py

@@ -1,7 +1,7 @@
-import autograd.numpy as anp
+import jax.numpy as jnp


logical ordering of imports

jhuggins · 2023-11-21T17:50:01Z

viabel/tests/test_convenience.py

@@ -1,7 +1,7 @@
-import autograd.numpy as anp
+import jax.numpy as jnp


why do you need to use Jax for everything here? seems only necessary if you are going to autodiff. Same question applies to other test files

jhuggins · 2023-11-21T17:51:28Z

viabel/tests/test_models.py

 def _test_model(m, x, supports_tempering, supports_constrain):
-    check_vjp(m, x)
-    check_vjp(m, x[0])
+    #check_vjp(m, (x,), modes=['rev'], order=2)


delete these comments?

jhuggins

Looking good but the tests aren't passing

avehtari · 2024-09-24T17:31:29Z

@CyrusZhang73 just checking whether you are still planning to work on this PR? I'm asking as @charlesm93 is interested in testing Viabel with BridgeStan.

CyrusZhang73 added 30 commits September 25, 2023 13:42

Add files via upload

aacbfc9

Update _distributions.py

1c225c7

Delete viabel/_mc_diagnostics.py

ed52ae2

Add files via upload

e660c48

Update _utils.py

9ddf26e

Update approximations.py

8af155e

Update models.py

6c52ee8

Update objectives.py

5aced28

Update objectives.py

7e80d31

Update optimization.py

3c42431

Update test_convenience.py

e2ad4b5

Update test_models.py

dfcf197

Add files via upload

497832d

Update test_objectives.py

54a4085

Update test_optimization.py

4426c96

Update _utils.py

4c3f78e

Delete viabel/jax_paragami directory

12bd16d

Add files via upload

e78c030

Update approximations.py

cd8dd1c

Update approximations.py

cf68eb7

Update _utils.py

97c3f79

Update objectives.py

e740013

Update objectives.py

e198151

Update objectives.py

87946b4

Update test_objectives.py

25fe10c

Update objectives.py

dc70fda

Update objectives.py

c56166f

Update requirements.txt

8c6ffd1

Update requirements-docs.txt

401032f

Update requirements.txt

bcad736

CyrusZhang73 added 23 commits October 18, 2023 15:40

Update test_objectives.py

957ead8

Update objectives.py

bdaba59

Update test_objectives.py

d83074d

Update approximations.py

28f63b6

Update approximations.py

e89c983

Update objectives.py

29ae60b

Update approximations.py

a4da3e1

Update objectives.py

0ea0e81

Update objectives.py

851e1c6

Update test_objectives.py

283a134

Update test_objectives.py

1efe233

Update approximations.py

b0d33b4

Update test_objectives.py

b4df4c9

Update test_optimization.py

8376b24

Update optimization.py

8de7241

Update optimization.py

3976e01

Update requirements.txt

eddf500

Update requirements-dev.txt

4f91c7d

Update requirements.txt

f3bb1ba

Update requirements-dev.txt

79d8279

Update requirements-dev.txt

37b5599

Update optimization.py

e0db797

Update test_optimization.py

2d2f9ca

jhuggins assigned CyrusZhang73 Nov 7, 2023

jhuggins requested changes Nov 21, 2023

View reviewed changes

CyrusZhang73 added 2 commits November 27, 2023 16:50

I modified most of them.

0264df0

change function value_and_grad()

8248b6f

jhuggins requested changes Feb 13, 2024

View reviewed changes



		def bbvi(dimension, *, n_iters=10000, num_mc_samples=10, log_density=None,
		def bbvi(dimension, *, n_iters=3000, num_mc_samples=10, log_density=None,



		def vi_diagnostics(var_param, *, objective=None, model=None, approx=None, n_samples=100000):
		def vi_diagnostics(var_param, *, objective=None, model=None, approx=None, n_samples=3000):

		@@ -1,7 +1,7 @@
		import autograd.numpy as anp
		import jax.numpy as jnp

Switching to Bridgestan and JAX #58

Are you sure you want to change the base?

Switching to Bridgestan and JAX #58

Uh oh!

Conversation

CyrusZhang73 commented Oct 17, 2023

Uh oh!

codecov-commenter commented Oct 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jhuggins left a comment

Choose a reason for hiding this comment

Uh oh!

avehtari commented Sep 24, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codecov-commenter commented Oct 18, 2023 •

edited

Loading