Use unit normal as default init_dist in GaussianRandomWalk and AR

RELEASE-NOTES.md

@@ -25,11 +25,14 @@ Also check out the [milestones](https://github.com/pymc-devs/pymc/milestones) fo
 
 All of the above apply to:
 
-Signature and default parameters changed for several distributions (see [#5628](https://github.com/pymc-devs/pymc/pull/5628)):
-    - `pm.StudentT` now requires either `sigma` or `lam` as kwarg
-    - `pm.StudentT` now requires `nu` to be specified (no longer defaults to 1)
-    - `pm.AsymmetricLaplace` positional arguments re-ordered
-    - `pm.AsymmetricLaplace` now requires `mu` to be specified (no longer defaults to 0)
+Signature and default parameters changed for several distributions:
+    - `pm.StudentT` now requires either `sigma` or `lam` as kwarg (see [#5628](https://github.com/pymc-devs/pymc/pull/5628))
+    - `pm.StudentT` now requires `nu` to be specified (no longer defaults to 1) (see [#5628](https://github.com/pymc-devs/pymc/pull/5628))
+    - `pm.AsymmetricLaplace` positional arguments re-ordered (see [#5628](https://github.com/pymc-devs/pymc/pull/5628))
+    - `pm.AsymmetricLaplace` now requires `mu` to be specified (no longer defaults to 0) (see [#5628](https://github.com/pymc-devs/pymc/pull/5628))
+    - `ZeroInflatedPoisson` `theta` parameter was renamed to `mu` (see [#5584](https://github.com/pymc-devs/pymc/pull/5584)).
+    - `pm.GaussianRandomWalk` initial distribution defaults to unit normal instead of flat (see[#5779](https://github.com/pymc-devs/pymc/pull/5779))
+    - `pm.AR` initial distribution defaults to unit normal instead of flat (see[#5779](https://github.com/pymc-devs/pymc/pull/5779))
 - BART was removed [#5566](https://github.com/pymc-devs/pymc/pull/5566). It is now available from [pymc-experimental](https://github.com/pymc-devs/pymc-experimental)
 - The `pm.EllipticalSlice` sampler was removed (see [#5756](https://github.com/pymc-devs/pymc/issues/5756)).
 - `BaseStochasticGradient` was removed (see [#5630](https://github.com/pymc-devs/pymc/pull/5630))
@@ -80,7 +83,7 @@ Signature and default parameters changed for several distributions (see [#5628](
   - The function `replace_with_values` function has been added to `gp.utils`.
   - `MarginalSparse` has been renamed `MarginalApprox`.
   - Removed `MixtureSameFamily`. `Mixture` is now capable of handling batched multivariate components (see [#5438](https://github.com/pymc-devs/pymc/pull/5438)).
-  - `ZeroInflatedPoisson` `theta` parameter was renamed to `mu` (see [#5584](https://github.com/pymc-devs/pymc/pull/5584)).
+  - Removed `AR1`, `AR` of order 1 should be used instead. (see [5734](https://github.com/pymc-devs/pymc/pull/5734)).
 - ...
 
 ### Expected breaks

pymc/distributions/timeseries.py

@@ -225,7 +225,7 @@ class GaussianRandomWalk(distribution.Continuous):
         sigma > 0, innovation standard deviation, defaults to 1.0
     init : unnamed distribution
         Univariate distribution of the initial value, created with the `.dist()` API.
-        Defaults to Normal with same `mu` and `sigma` as the GaussianRandomWalk
+        Defaults to a unit Normal.
 
         .. warning:: init will be cloned, rendering them independent of the ones passed as input.
 
@@ -265,7 +265,7 @@ def dist(
 
         # If no scalar distribution is passed then initialize with a Normal of same mu and sigma
         if init is None:
-            init = Normal.dist(mu, sigma)
+            init = Normal.dist(0, 1)
         else:
             if not (
                 isinstance(init, at.TensorVariable)
@@ -361,7 +361,7 @@ class AR(SymbolicDistribution):
         Whether the first element of rho should be used as a constant term in the AR
         process. Defaults to False
     init_dist: unnamed distribution, optional
-        Scalar or vector distribution for initial values. Defaults to Normal(0, sigma).
+        Scalar or vector distribution for initial values. Defaults to a unit Normal.
         Distribution should be  created via the `.dist()` API, and have  dimension
         (*size, ar_order). If not, it will be automatically resized.
 
@@ -452,8 +452,7 @@ def dist(
                     f"got ndim_supp={init_dist.owner.op.ndim_supp}.",
                 )
         else:
-            # Sigma must broadcast with ar_order
-            init_dist = Normal.dist(sigma=at.shape_padright(sigma), size=(*sigma.shape, ar_order))
+            init_dist = Normal.dist(0, 1, size=(*sigma.shape, ar_order))
 
         # Tell Aeppl to ignore init_dist, as it will be accounted for in the logp term
         init_dist = ignore_logprob(init_dist)

pymc/tests/test_distributions.py

@@ -2610,7 +2610,7 @@ def test_gaussianrandomwalk(self):
         def ref_logp(value, mu, sigma, steps):
             # Relying on fact that init will be normal by default
             return (
-                scipy.stats.norm.logpdf(value[0], mu, sigma)
+                scipy.stats.norm.logpdf(value[0])
                 + scipy.stats.norm.logpdf(np.diff(value), mu, sigma).sum()
             )
 

pymc/tests/test_distributions_timeseries.py

@@ -333,6 +333,7 @@ def test_batched_sigma(self):
                 "y",
                 beta_tp,
                 sigma=sigma,
+                init_dist=Normal.dist(0, sigma[..., None]),
                 size=batch_size,
                 steps=steps,
                 initval=y_tp,
@@ -346,6 +347,7 @@ def test_batched_sigma(self):
                         f"y_{i}{j}",
                         beta_tp,
                         sigma=sigma[i][j],
+                        init_dist=Normal.dist(0, sigma[i][j]),
                         shape=steps,
                         initval=y_tp[i, j],
                         ar_order=ar_order,
@@ -371,7 +373,7 @@ def test_batched_init_dist(self):
         beta_tp = aesara.shared(np.random.randn(ar_order), shape=(3,))
         y_tp = np.random.randn(batch_size, steps)
         with Model() as t0:
-            init_dist = Normal.dist(0.0, 0.01, size=(batch_size, ar_order))
+            init_dist = Normal.dist(0.0, 1.0, size=(batch_size, ar_order))
             AR("y", beta_tp, sigma=0.01, init_dist=init_dist, steps=steps, initval=y_tp)
         with Model() as t1:
             for i in range(batch_size):