Separate iid from set x by Jocho-Smith · Pull Request #1945 · sbi-dev/sbi

Jocho-Smith · 2026-07-22T07:51:56Z

This branch builds upon #1943 . There I build bind() that internally still users set_x for setting _x_o and x_is_iid. This PR is about resolving this dependence entirely and removing set_x from inside bind.

3 things: remove set_x from BasePotential.init; replace overlooked set_x inside vi_divergence; remove set_x usage inside bind()

This PR experiments with the next step for the GSoC project: decouple iid from set_x.

It requires #1943 to be merged, which is not the case yet so, I took the commits from there with me.

more description is coming

- Add bind() method to BasePotential (abstract) and CustomPotentialWrapper - Add bind() to LikelihoodBasedPotential, PosteriorBasedPotential, RatioBasedPotential - Add bind() to VectorFieldBasedPotential with full parameter support - Add bind() to EnsemblePotential in ensemble_posterior.py - Add bind() to ConditionedPotential in conditional_density_utils.py - Fix _base_recursor to use local variable _active_holder to avoid Python scoping issues with the _active parameter - Update posteriors to use bind() instead of set_x() for immutability - Fix PytorchReturnTypeWrapper, MultipleIndependent, OneDimPriorWrapper init order and add __deepcopy__ methods for proper deepcopy support - Update error messages to reference bind() instead of set_x() - Add bind() to test FakePotential and TractablePotential classes

- Add bind() method to BasePotential (abstract) and CustomPotentialWrapper - Add bind() to LikelihoodBasedPotential, PosteriorBasedPotential, RatioBasedPotential - Add bind() to VectorFieldBasedPotential with full parameter support - Add bind() to EnsemblePotential in ensemble_posterior.py - Add bind() to ConditionedPotential in conditional_density_utils.py - Fix _base_recursor to use local variable _active_holder to avoid Python scoping issues with the _active parameter - Update posteriors to use bind() instead of set_x() for immutability - Fix PytorchReturnTypeWrapper, MultipleIndependent, OneDimPriorWrapper init order and add __deepcopy__ methods for proper deepcopy support - Update error messages to reference bind() instead of set_x() - Add bind() to test FakePotential and TractablePotential classes - Replace remaining set_x calls in VectorFieldPosterior with bind

…ind-method-clean

…d set_x inside vi_divergence; remove set_x usage inside bind()

…ibute 'guidance_method'

Jocho-Smith · 2026-07-22T09:15:01Z

all tests passing, good! Now I'll work on merging #1943 and then merge this next.

Since this one already decouples set_x and bind I can continue preparing the deprecation here: #1946

codecov · 2026-07-22T10:52:23Z

Codecov Report

❌ Patch coverage is 85.71429% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 87.91%. Comparing base (587f922) to head (caa994d).
⚠️ Report is 40 commits behind head on gsoc-2026.

Files with missing lines	Patch %	Lines
.../inference/potentials/posterior_based_potential.py	50.00%	4 Missing ⚠️
sbi/inference/potentials/vector_field_potential.py	87.50%	2 Missing ⚠️
sbi/samplers/vi/vi_divergence_optimizers.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##           gsoc-2026    #1945      +/-   ##
=============================================
- Coverage      87.99%   87.91%   -0.09%     
=============================================
  Files            144      144              
  Lines          13499    13756     +257     
=============================================
+ Hits           11879    12094     +215     
- Misses          1620     1662      +42

Flag	Coverage Δ
fast	`81.62% <85.71%> (?)`

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
sbi/inference/posteriors/vi_posterior.py	`78.60% <100.00%> (+0.04%)`	⬆️
sbi/inference/potentials/base_potential.py	`91.66% <100.00%> (-1.20%)`	⬇️
...inference/potentials/likelihood_based_potential.py	`83.33% <100.00%> (+1.22%)`	⬆️
sbi/inference/potentials/ratio_based_potential.py	`100.00% <100.00%> (ø)`
sbi/utils/torchutils.py	`70.53% <100.00%> (+2.33%)`	⬆️
sbi/utils/user_input_checks_utils.py	`89.06% <100.00%> (-1.05%)`	⬇️
sbi/samplers/vi/vi_divergence_optimizers.py	`84.69% <50.00%> (-0.33%)`	⬇️
sbi/inference/potentials/vector_field_potential.py	`90.84% <87.50%> (-0.14%)`	⬇️
.../inference/potentials/posterior_based_potential.py	`90.76% <50.00%> (-5.60%)`	⬇️

... and 2 files with indirect coverage changes

Resolved conflicts: - likelihood_based_potential.py: keep HEAD (direct attribute set) - posterior_based_potential.py: keep HEAD (direct attribute set) - ratio_based_potential.py: keep HEAD (direct attribute set) - vector_field_potential.py: keep HEAD (direct attribute set + guidance_method) - conditional_density_utils.py: accept upstream (fixes _x_is_iid on bound)

dgedon · 2026-07-22T20:24:14Z

    def set_x(self, x_o: Optional[Tensor], x_is_iid: Optional[bool] = True):
        """Check the shape of the observed data and, if valid, set it."""
        if x_o is not None:
            x_o = process_x(x_o).to(self.device)
        self._x_o = x_o
        self._x_is_iid = x_is_iid


Shouldn't we remove this part in this PR already?

Ah no, we add the deprecation warning in #1946

dgedon · 2026-07-23T07:00:06Z


    def bind(self, x_o: Tensor, x_is_iid: bool = True) -> "LikelihoodBasedPotential":
        """Create new potential with x bound, without mutable state."""
+        from sbi.utils.user_input_checks import process_x


we should import that not within the function but at file level

dgedon · 2026-07-23T07:01:14Z


    def bind(self, x_o: Tensor, x_is_iid: bool = True) -> "RatioBasedPotential":
        """Create new potential with x bound, without mutable state."""
+        from sbi.utils.user_input_checks import process_x


same is in the likelihoood based potential

dgedon · 2026-07-23T07:01:17Z


    def bind(self, x_o: Tensor, x_is_iid: bool = False) -> "PosteriorBasedPotential":
        """Create new potential with x bound, without mutable state."""
+        from sbi.utils.user_input_checks import process_x


same is in the likelihoood based potential

dgedon · 2026-07-23T07:22:35Z

        """
-        super().set_x(x_o, x_is_iid)
+        if x_o is not None:
+            from sbi.utils.user_input_checks import process_x


Import on file level

dgedon · 2026-07-23T07:24:53Z

            device=self.device,
-            iid_method=self.iid_method,
-            iid_params=self.iid_params,
+            iid_method=iid_method or self.iid_method,


this should proabbly also be a id_method if iid_method is not None else self.iid_method,

dgedon · 2026-07-23T07:32:05Z

-            validate_args=(
-                prior._validate_args if validate_args is None else validate_args
-            ),
+            validate_args=False,


flag (see file level comment)

dgedon · 2026-07-23T07:32:10Z

+    def __deepcopy__(self, memo):
+        """Ensure prior attribute is preserved during deepcopy."""
+        cls = self.__class__
+        result = cls.__new__(cls)
+        memo[id(self)] = result
+        for k, v in self.__dict__.items():
+            setattr(result, k, deepcopy(v, memo))
+        return result
+


flag (see file level comment)

dgedon · 2026-07-23T07:32:31Z

Here we seem to have the same modifications that we dicussed in #1943 before.

there is code duplication for __deepcopy__

I don't see where deepcopy is used and why it is necessary to introduce

we silently change the validate_args hard coded to False which seems to be a bug.

ouh sorry. I messed up the merge. Will fix it.

dgedon · 2026-07-23T07:38:38Z

+        if x_o is not None:
+            x_o = process_x(x_o).to(self.device)
+        self._x_o = x_o
+        self._x_is_iid = True


compared with set_x() we hard to _x_is_iid to True here. Since it's not a method argument, maybe True is a good default. We have to nesure though that it is set correctly in each inherited class. Because right now by calling super() in e.g. the likelihood potential, we will ahve x_is_iid=True by default

right now by calling super() in e.g. the likelihood potential, we will ahve x_is_iid=True by default

This is exactly what is happening right now on the main branch. likelihood calls super without _x_is_iid : https://github.com/sbi-dev/sbi/blob/main/sbi/inference/potentials/likelihood_based_potential.py#L80

BasePotential.init always calls set_x without any iid information: https://github.com/sbi-dev/sbi/blob/main/sbi/inference/potentials/base_potential.py#L33

x_is_iid is set with default value True: https://github.com/sbi-dev/sbi/blob/main/sbi/inference/potentials/base_potential.py#L60

…_method for VectorFieldBasedPotential

Jocho-Smith and others added 10 commits July 22, 2026 07:33

Merge branch 'bind-method-clean' of github.com:jocho-smith/sbi into b…

30459d9

…ind-method-clean

fix attempt with bound.neural_ode.params.update

8fdc10c

checkpoint after full bind replacement of set_x

968b58b

Merge branch 'gsoc-2026' into bind-method-clean

84c4731

fixup:slightly wrong last merge

ee2adfa

3 things: remove set_x from BasePotential.__init__; replace overlooke…

0fe9359

…d set_x inside vi_divergence; remove set_x usage inside bind()

fix attempt: VI training broken due to bind

e8444c5

fixup: AttributeError: 'VectorFieldBasedPotential' object has no attr…

81b0709

…ibute 'guidance_method'

Jocho-Smith mentioned this pull request Jul 22, 2026

Add .bind() method to potentials #1943

Merged

Jocho-Smith mentioned this pull request Jul 22, 2026

Test set x deprecation #1946

Merged

This was referenced Jul 22, 2026

migrated tests from set_x to bind #1948

Merged

GSoC2026: Project 3 dashboard #1950

Open

dgedon reviewed Jul 23, 2026

View reviewed changes

reverted user_input_checks_utils changes; moved imports up; fixed iid…

caa994d

…_method for VectorFieldBasedPotential

Jocho-Smith requested a review from dgedon July 23, 2026 12:02

dgedon approved these changes Jul 23, 2026

View reviewed changes

dgedon merged commit 88585d8 into sbi-dev:gsoc-2026 Jul 23, 2026
17 checks passed

Uh oh!

Conversation

Jocho-Smith commented Jul 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Jocho-Smith commented Jul 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Jul 22, 2026 • 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Jocho-Smith Jul 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Jocho-Smith commented Jul 22, 2026 •

edited

Loading

Jocho-Smith commented Jul 22, 2026 •

edited

Loading

codecov Bot commented Jul 22, 2026 •

edited

Loading

Jocho-Smith Jul 23, 2026 •

edited

Loading