Merge branch 'main' of github.com:pymc-devs/pymc into deprecate-generator

Author: williambdean

conda-envs/environment-dev.yml

@@ -13,7 +13,7 @@ dependencies:
 - numpy>=1.25.0
 - pandas>=0.24.0
 - pip
-- pytensor>=2.26.2,<2.27
+- pytensor>=2.26.2,<2.28
 - python-graphviz
 - networkx
 - scipy>=1.4.1

conda-envs/environment-docs.yml

@@ -11,7 +11,7 @@ dependencies:
 - numpy>=1.25.0
 - pandas>=0.24.0
 - pip
-- pytensor>=2.26.2,<2.27
+- pytensor>=2.26.2,<2.28
 - python-graphviz
 - rich>=13.7.1
 - scipy>=1.4.1

conda-envs/environment-jax.yml

@@ -21,7 +21,7 @@ dependencies:
 - numpyro>=0.8.0
 - pandas>=0.24.0
 - pip
-- pytensor>=2.26.2,<2.27
+- pytensor>=2.26.2,<2.28
 - python-graphviz
 - networkx
 - rich>=13.7.1

conda-envs/environment-test.yml

@@ -14,7 +14,7 @@ dependencies:
 - numpy>=1.25.0
 - pandas>=0.24.0
 - pip
-- pytensor>=2.26.2,<2.27
+- pytensor>=2.26.2,<2.28
 - python-graphviz
 - networkx
 - rich>=13.7.1

conda-envs/windows-environment-dev.yml

@@ -13,7 +13,7 @@ dependencies:
 - numpy>=1.25.0
 - pandas>=0.24.0
 - pip
-- pytensor>=2.26.2,<2.27
+- pytensor>=2.26.2,<2.28
 - python-graphviz
 - networkx
 - rich>=13.7.1

conda-envs/windows-environment-test.yml

@@ -16,7 +16,7 @@ dependencies:
 - numpy>=1.25.0
 - pandas>=0.24.0
 - pip
-- pytensor>=2.26.2,<2.27
+- pytensor>=2.26.2,<2.28
 - python-graphviz
 - networkx
 - rich>=13.7.1

docs/source/guides/Probability_Distributions.rst

@@ -29,21 +29,32 @@ A variable requires at least a ``name`` argument, and zero or more model paramet
 
 Probability distributions are all subclasses of ``Distribution``, which in turn has two major subclasses: ``Discrete`` and ``Continuous``. In terms of data types, a ``Continuous`` random variable is given whichever floating point type is defined by ``pytensor.config.floatX``, while ``Discrete`` variables are given ``int16`` types when ``pytensor.config.floatX`` is ``float32``, and ``int64`` otherwise.
 
-All distributions in ``pm.distributions`` will have two important methods: ``random()`` and ``logp()`` with the following signatures:
+All distributions in ``pm.distributions`` are associated with two key functions:
+
+1. ``logp(dist, value)`` - Calculates log-probability at given value
+2. ``draw(dist, size=...)`` - Generates random samples
+
+For example, with a normal distribution:
 
 ::
 
-    class SomeDistribution(Continuous):
+    with pm.Model():
+        x = pm.Normal('x', mu=0, sigma=1)
+
+    # Calculate log-probability
+    log_prob = pm.logp(x, 0.5)
+
+    # Generate samples
+    samples = pm.draw(x, size=100)
 
-        def random(self, point=None, size=None):
-            ...
-            return random_samples
+Custom distributions using ``CustomDist`` should provide logp via the ``dist`` parameter:
+
+::
 
-        def logp(self, value):
-            ...
-            return total_log_prob
+    def custom_logp(value, mu):
+        return -0.5 * (value - mu)**2
 
-PyMC expects the ``logp()`` method to return a log-probability evaluated at the passed ``value`` argument. This method is used internally by all of the inference methods to calculate the model log-probability that is used for fitting models. The ``random()`` method is used to simulate values from the variable, and is used internally for posterior predictive checks.
+    custom_dist = pm.CustomDist('custom', dist=custom_logp, mu=0)
 
 
 Custom distributions
@@ -58,7 +69,7 @@ An exponential survival function, where :math:`c=0` denotes failure (or non-surv
     f(c, t) = \left\{ \begin{array}{l} \exp(-\lambda t), \text{if c=1} \\
                \lambda \exp(-\lambda t), \text{if c=0}  \end{array} \right.
 
-Such a function can be implemented as a PyMC distribution by writing a function that specifies the log-probability, then passing that function as a keyword argument to the ``DensityDist`` function, which creates an instance of a PyMC distribution with the custom function as its log-probability.
+Such a function can be implemented as a PyMC distribution by writing a function that specifies the log-probability, then passing that function as a keyword argument to the ``CustomDist`` function, which creates an instance of a PyMC distribution with the custom function as its log-probability.
 
 For the exponential survival function, this is:
 
@@ -67,7 +78,7 @@ For the exponential survival function, this is:
     def logp(value, t, lam):
         return (value * log(lam) - lam * t).sum()
 
-    exp_surv = pm.DensityDist('exp_surv', t, lam, logp=logp, observed=failure)
+    exp_surv = pm.CustomDist('exp_surv', dist=logp, t=t, lam=lam, observed=failure)
 
 Similarly, if a random number generator is required, a function returning random numbers corresponding to the probability distribution can be passed as the ``random`` argument.
 
@@ -98,10 +109,10 @@ This allows for probabilities to be calculated and random numbers to be drawn.
 
 ::
 
-    >>> y.logp(4).eval()
+    >>> pm.logp(y, 4).eval()
     array(-1.5843639373779297, dtype=float32)
 
-    >>> y.random(size=3)
+    >>> pm.draw(y, size=3)
     array([5, 4, 3])
 
 

requirements-dev.txt

@@ -17,7 +17,7 @@ numpydoc
 pandas>=0.24.0
 polyagamma
 pre-commit>=2.8.0
-pytensor>=2.26.2,<2.27
+pytensor>=2.26.2,<2.28
 pytest-cov>=2.5
 pytest>=3.0
 rich>=13.7.1

requirements.txt

@@ -3,7 +3,7 @@ cachetools>=4.2.1
 cloudpickle
 numpy>=1.25.0
 pandas>=0.24.0
-pytensor>=2.26.1,<2.27
+pytensor>=2.26.1,<2.28
 rich>=13.7.1
 scipy>=1.4.1
 threadpoolctl>=3.1.0,<4.0.0

tests/test_data.py

@@ -373,7 +373,7 @@ def test_implicit_coords_series(self, seeded_test):
         pd = pytest.importorskip("pandas")
         ser_sales = pd.Series(
             data=np.random.randint(low=0, high=30, size=22),
-            index=pd.date_range(start="2020-05-01", periods=22, freq="24H", name="date"),
+            index=pd.date_range(start="2020-05-01", periods=22, freq="24h", name="date"),
             name="sales",
         )
         with pm.Model() as pmodel: