Remove `k_endog` & `k_exog` parameters in SSM #599

Dekermanjian · 2025-10-19T20:15:39Z

This PR is related to #589 and aims to remove redundant parameters k_endog and k_exog from SSM models. The following models will be updated:

In addition, this PR will also address

VARMAX input validation functionality should be moved out to a helper #587

Closes #587
Closes #589

Dekermanjian · 2025-10-25T11:57:40Z

Hey @jessegrabowski, after updating all the models and the regression component in the structural api the complexity of the validation code dropped quite a bit. I am looking at pulling out that VARMAX validation utility to use across all the other models (SARIMAX, ETS, DFM) but the only commonality between them is how the endog_names are handled. I can pull out the validation for endog_names and wrap that up in a utility but that is only 4 lines of code. ~~Do you think it is worth it?~~ Never mind, I am going to do it. Once we have more models it will be worth it.

jessegrabowski · 2025-10-27T01:06:05Z

I was going to say yes anyway :D

jessegrabowski · 2025-10-27T12:33:24Z

I am looking at pulling out that VARMAX validation utility to use across all the other models

Mark that this PR will close both #589 and #587 as a result

Dekermanjian · 2025-10-27T21:29:57Z

Jesse, I marked the two open issues for closer upon merging this PR.

jessegrabowski

Nice work as always! Left some feedback.

Also rebase, we just merged the Pytensor update that will fix tests.

jessegrabowski · 2025-10-29T01:31:37Z

pymc_extras/statespace/models/structural/components/regression.py

    def __init__(
        self,
-        k_exog: int | None = None,
+        # k_exog: int | None = None,


Delete, no need to leave flotsam in the code.

My apologies, I missed that. I will delete it in the next push.

jessegrabowski · 2025-10-29T01:34:27Z

pymc_extras/statespace/models/utilities.py

    return exog_dims
+
+
+def _validate_endog_names(endog_names) -> int:


Change this to validate_names, and add a optional: bool =True argument so it can either error or return None. Then you can eliminate _get_state_names in the regression component and use this instead. Also can use it for the exog names in SARIMAX, etc.

I added a var_name argument as well so that if it does raise a value error the message will clearly state what variable is missing.

jessegrabowski · 2025-10-29T01:37:44Z

tests/statespace/models/test_SARIMAX.py


+@pytest.mark.filterwarnings(
+    "ignore::RuntimeWarning"
+)  # Needed this due to RuntimeWarning: divide by zero encountered in matmul


Are you on a Mac? This is a known bug with Mx chips, see here: numpy/numpy#29820

Yes, I am on a Mac. Should I run in x86 using rosetta instead until they fix the issue?

No its fine, it doesn't actually do anything. But you don't need to filterwarnings either. You can just do it locally for your own testing

Okay, that is reasonable. I removed the filter before pushing new commits.

jessegrabowski · 2025-10-29T01:38:02Z

tests/statespace/models/test_VARMAX.py

+    # def test_create_varmax_with_exogenous_k_exog_int(self, data):
+    #     mod = BayesianVARMAX(
+    #         endog_names=["realgdp", "realcons", "realinv"],
+    #         order=(1, 0),
+    #         exog_state_names=["exogenous_0", "exogenous_1"],
+    #         verbose=False,
+    #         measurement_error=False,
+    #         stationary_initialization=False,
+    #     )
+    #     assert mod.k_exog == 2
+    #     assert mod.exog_state_names == ["exogenous_0", "exogenous_1"]
+    #     assert mod.data_names == ["exogenous_data"]
+    #     assert mod.param_dims["beta_exog"] == ("observed_state", "exogenous")
+    #     assert mod.coords["exogenous"] == ["exogenous_0", "exogenous_1"]
+    #     assert mod.param_info["beta_exog"]["shape"] == (mod.k_endog, 2)
+    #     assert mod.param_info["beta_exog"]["dims"] == ("observed_state", "exogenous")


(also applies to other commented out blocks below)

Yup, will do! Didn't want to delete anything before you approve of it.

jessegrabowski · 2025-10-29T01:38:43Z

tests/statespace/models/test_VARMAX.py

            }
        else:
-            exog_names = exog_state_names or [f"exogenous_{i}" for i in range(k_exog)]
+            exog_names = exog_state_names


I might be misunderstanding your question, but I removed the or statement because there no longer is a k_exog parameter.

I meant why keep this, since all it does is rename a variable. There should be a more elegant way

Yes, you are absolutely right. That is silly. I changed it to directly pass in exog_state_names.

… and updated tests accordingly

…exogenous variables and updated tests accordingly

…dog_names required and exog_state_names required for exogenous variables and updated tests accordingly

…quired and exog_names required if exogenous variables are requested and updated tests accordingly

…ingly

…k_endog and k_exog parameters

…eused in both endog and some exog cases

jessegrabowski requested changes Oct 29, 2025

View reviewed changes

Dekermanjian added 8 commits October 29, 2025 16:57

removed k_endog argument from BayesianETS making endog_names required…

3646089

… and updated tests accordingly

removed k_exog from SARIMAX preferring exog_state_names for defining …

f4b105f

…exogenous variables and updated tests accordingly

Updated VARMAX model by removing k_endog & k_exog arguments making en…

59c8b10

…dog_names required and exog_state_names required for exogenous variables and updated tests accordingly

updated DFM by removing k_endog & k_exog args and made endog_names re…

b63c381

…quired and exog_names required if exogenous variables are requested and updated tests accordingly

removed k_exog from STS regression component and updated tests accord…

4fab3fc

…ingly

updated docstrings of VARMAX and DFM to relect changes in removal of …

9ecd1af

…k_endog and k_exog parameters

moved endog_names validation into stand alone utility

90f2a78

removed commented code and tests, updated validate names test to be r…

330dc19

…eused in both endog and some exog cases

Dekermanjian force-pushed the remove_kendog_kexog_ssm branch from 9a3e24d to 330dc19 Compare October 29, 2025 23:12

		return exog_dims


		def _validate_endog_names(endog_names) -> int:

Remove k_endog & k_exog parameters in SSM #599

Are you sure you want to change the base?

Remove k_endog & k_exog parameters in SSM #599

Uh oh!

Conversation

Dekermanjian commented Oct 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Dekermanjian commented Oct 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jessegrabowski commented Oct 27, 2025

Uh oh!

jessegrabowski commented Oct 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Dekermanjian commented Oct 27, 2025

Uh oh!

jessegrabowski left a comment

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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Remove `k_endog` & `k_exog` parameters in SSM #599

Remove `k_endog` & `k_exog` parameters in SSM #599

Dekermanjian commented Oct 19, 2025 •

edited

Loading

Dekermanjian commented Oct 25, 2025 •

edited

Loading

jessegrabowski commented Oct 27, 2025 •

edited

Loading