Implemented Generalized Gamma RV

aesara/tensor/random/basic.py

@@ -569,6 +569,26 @@ class BetaBinomialRV(ScipyRandomVariable):
 betabinom = BetaBinomialRV()
 
 
+class GenGammaRV(ScipyRandomVariable):
+    name = "gengamma"
+    ndim_supp = 0
+    ndims_params = [0, 0, 0]
+    dtype = "floatX"
+    _print_name = ("GG", "\\operatorname{GG}")
+
+    def __call__(self, alpha=1.0, p=1.0, lambd=1.0, size=None, **kwargs):
+        return super().__call__(alpha, p, lambd, size=size, **kwargs)
+
+    @classmethod
+    def rng_fn_scipy(cls, rng, alpha, p, lambd, size):
+        return stats.gengamma.rvs(
+            alpha / p, p, scale=lambd, size=size, random_state=rng
+        )
+
+
+gengamma = GenGammaRV()
+
+
 class MultinomialRV(RandomVariable):
     """A Multinomial random variable type.
 
@@ -794,4 +814,5 @@ __all__ = [
     "uniform",
     "standard_normal",
     "negative_binomial",
+    "gengamma",
 ]

tests/tensor/random/test_basic.py

@@ -28,6 +28,7 @@ from aesara.tensor.random.basic import (
     dirichlet,
     exponential,
     gamma,
+    gengamma,
     geometric,
     gumbel,
     halfcauchy,
@@ -1093,6 +1094,48 @@ def test_betabinom_samples(M, a, p, size):
     )
 
 
+@pytest.mark.parametrize(
+    "alpha, p, lambd, size",
+    [
+        (
+            np.array(2, dtype=config.floatX),
+            np.array(3, dtype=config.floatX),
+            np.array(5, dtype=config.floatX),
+            None,
+        ),
+        (
+            np.array(1, dtype=config.floatX),
+            np.array(1, dtype=config.floatX),
+            np.array(10, dtype=config.floatX),
+            [],
+        ),
+        (
+            np.array(2, dtype=config.floatX),
+            np.array(2, dtype=config.floatX),
+            np.array(10, dtype=config.floatX),
+            [2, 3],
+        ),
+        (
+            np.full((1, 2), 2, dtype=config.floatX),
+            np.array(2, dtype=config.floatX),
+            np.array(10, dtype=config.floatX),
+            None,
+        ),
+    ],
+)
+def test_gengamma_samples(alpha, p, lambd, size):
+    compare_sample_values(
+        gengamma,
+        alpha,
+        p,
+        lambd,
+        size=size,
+        test_fn=lambda *args, size=None, random_state=None, **kwargs: gengamma.rng_fn(
+            random_state, *(args + (size,))
+        ),
+    )
+
+
 @pytest.mark.parametrize(
     "M, p, size, test_fn",
     [