Refactor get_tau_sigma to handle lists (#5747)

* Refactor get_tau_sigma to handle lists

* Fixed syntax error

* Black formatted

* Change assertions to ValueError raises

* Prior predictions constant data (#5723)

* support return constant_data with prior pred

* pre-commit format

* added test

* code review

* Fix issue probably-meant-fstring found at https://codereview.doctor (#5726)

* Add coords argument to pymc.set_data (#5588)


Co-authored-by: Oriol (ZBook) <[email protected]>

* remove MultinomialRV override

* ⬆️ UPGRADE: Autoupdate pre-commit config

* Group GaussianRandomWalk tests in single class

* Infer steps from shape in GaussianRandomWalk

* Unpin upper limit on numpydoc

Closes #5401

* ⬆️ UPGRADE: Autoupdate pre-commit config

* Standardize docstrings of input dists arguments and add warning about cloning

* Remove unnecessary tag in Simulator logp

* Replace deprecated tag.ignore_logprob

* Group compile_pymc tests in own class

* Remove remaining uses of default_updates in codebase

* Remove redundant/wrong docstrings from GaussianRandomWalk logp

* Add moment to GaussianRandomWalk and fix mu/sigma broadcasting bug

* Do not create temporary SymbolicDistribution just to retrieve number of RNGs needed

Reordered methods for consistency

* Move SymbolicDistribution docstring to body of class

* Refactor AR distribution

* Deprecates AR1 distribution
* Implements random and moment methods
* Batching works on the left as with other distributions

* Rename `pandas_to_array` to `convert_observed_data`

* Obtain step information from dims and observed

* Make AR steps extend shape beyond initial_dist

This is consistent with the meaning of steps in the GaussianRandomWalk and translates directly to the number of scan steps taken

Co-authored-by: TimOliverMaier <[email protected]>
Co-authored-by: code-review-doctor <[email protected]>
Co-authored-by: Somasree Majumder <[email protected]>
Co-authored-by: Oriol (ZBook) <[email protected]>
Co-authored-by: danhphan <[email protected]>
Co-authored-by: twiecki <[email protected]>
Co-authored-by: Ricardo <[email protected]>
Co-authored-by: Ravin Kumar <[email protected]>
Co-authored-by: Michael Osthege <[email protected]>

Author: 10 people

pymc/distributions/continuous.py

@@ -226,8 +226,9 @@ def get_tau_sigma(tau=None, sigma=None):
             if isinstance(sigma, Variable):
                 sigma_ = check_parameters(sigma, sigma > 0, msg="sigma > 0")
             else:
-                assert np.all(np.asarray(sigma) > 0)
-                sigma_ = sigma
+                sigma_ = np.asarray(sigma)
+                if np.any(sigma_ <= 0):
+                    raise ValueError("sigma must be positive")
             tau = sigma_**-2.0
 
     else:
@@ -237,9 +238,9 @@ def get_tau_sigma(tau=None, sigma=None):
             if isinstance(tau, Variable):
                 tau_ = check_parameters(tau, tau > 0, msg="tau > 0")
             else:
-                assert np.all(np.asarray(tau) > 0)
-                tau_ = tau
-
+                tau_ = np.asarray(tau)
+                if np.any(tau_ <= 0):
+                    raise ValueError("tau must be positive")
             sigma = tau_**-0.5
 
     return floatX(tau), floatX(sigma)

pymc/tests/test_distributions.py

@@ -2399,6 +2399,11 @@ def test_get_tau_sigma(self):
         with pytest.raises(ParameterValueError):
             sigma.eval()
 
+        sigma = [1, 2]
+        assert_almost_equal(
+            get_tau_sigma(sigma=sigma), [1.0 / np.array(sigma) ** 2, np.array(sigma)]
+        )
+
     @pytest.mark.parametrize(
         "value,mu,sigma,nu,logp",
         [