Update from comments

Author: ColCarroll

pymc3/distributions/distribution.py

@@ -476,10 +476,8 @@ def generate_samples(generator, *args, **kwargs):
             samples = generator(size=broadcast_shape, *args, **kwargs)
         elif dist_shape == broadcast_shape:
             samples = generator(size=size_tup + dist_shape, *args, **kwargs)
-        elif size_tup[-len(broadcast_shape):] != broadcast_shape:
-            samples = generator(size=size_tup + broadcast_shape, *args, **kwargs)
         else:
-            samples = generator(size=size_tup + dist_shape, *args, **kwargs)
+            samples = None
     # Args have been broadcast correctly, can just ask for the right shape out
     elif dist_shape[-len(broadcast_shape):] == broadcast_shape:
         samples = generator(size=size_tup + dist_shape, *args, **kwargs)
@@ -489,6 +487,9 @@ def generate_samples(generator, *args, **kwargs):
         samples = [generator(*args, **kwargs).reshape(size_tup + (1,)) for _ in range(np.prod(suffix, dtype=int))]
         samples = np.hstack(samples).reshape(size_tup + suffix)
     else:
+        samples = None
+
+    if samples is None:
         raise TypeError('''Attempted to generate values with incompatible shapes:
             size: {size}
             dist_shape: {dist_shape}

pymc3/model.py

@@ -1457,6 +1457,8 @@ def __init__(self, type=None, owner=None, index=None, name=None,
         if distribution is not None:
             self.model = model
             self.distribution = distribution
+            self.dshape = tuple(distribution.shape)
+            self.dsize = int(np.prod(distribution.shape))
 
             transformed_name = get_transformed_name(name, transform)
 

pymc3/sampling.py

@@ -1277,13 +1277,13 @@ def sample_ppc_w(traces, samples=None, models=None, weights=None,
     return {k: np.asarray(v) for k, v in ppc.items()}
 
 
-def sample_generative(samples=500, model=None, vars=None, random_seed=None):
-    """Generate samples from the generative model.
+def sample_prior_predictive(samples=500, model=None, vars=None, random_seed=None):
+    """Generate samples from the prior_predictive distribution.
 
     Parameters
     ----------
     samples : int
-        Number of samples from the prior to generate. Defaults to 500.
+        Number of samples from the prior predictive to generate. Defaults to 500.
     model : Model (optional if in `with` context)
     vars : iterable
         Variables for which to compute the posterior predictive samples.

pymc3/tests/test_distributions.py

@@ -909,10 +909,15 @@ def test_multinomial_mode(self, p, n):
         [[.25, .25, .25, .25], (1, 4), 3],
         # 3: expect to fail
         # [[.25, .25, .25, .25], (10, 4)],
+        [[.25, .25, .25, .25], (10, 1, 4), 5],
         # 5: expect to fail
         # [[[.25, .25, .25, .25]], (2, 4), [7, 11]],
         [[[.25, .25, .25, .25],
          [.25, .25, .25, .25]], (2, 4), 13],
+        [[[.25, .25, .25, .25],
+         [.25, .25, .25, .25]], (1, 2, 4), [23, 29]],
+        [[[.25, .25, .25, .25],
+         [.25, .25, .25, .25]], (10, 2, 4), [31, 37]],
         [[[.25, .25, .25, .25],
          [.25, .25, .25, .25]], (2, 4), [17, 19]],
     ])

pymc3/tests/test_sampling.py

@@ -348,7 +348,7 @@ def test_ignores_observed(self):
             positive_mu = pm.Deterministic('positive_mu', np.abs(mu))
             z = -1 - positive_mu
             pm.Normal('x_obs', mu=z, sd=1, observed=observed)
-            prior = pm.sample_generative()
+            prior = pm.sample_prior_predictive()
 
         assert (prior['mu'] < 90).all()
         assert (prior['positive_mu'] > 90).all()
@@ -360,15 +360,15 @@ def test_respects_shape(self):
             with pm.Model():
                 mu = pm.Gamma('mu', 3, 1, shape=1)
                 goals = pm.Poisson('goals', mu, shape=shape)
-                trace = pm.sample_generative(10)
+                trace = pm.sample_prior_predictive(10)
             if shape == 2:  # want to test shape as an int
                 shape = (2,)
             assert trace['goals'].shape == (10,) + shape
 
     def test_multivariate(self):
         with pm.Model():
             m = pm.Multinomial('m', n=5, p=np.array([0.25, 0.25, 0.25, 0.25]), shape=4)
-            trace = pm.sample_generative(10)
+            trace = pm.sample_prior_predictive(10)
 
         assert m.random(size=10).shape == (10, 4)
         assert trace['m'].shape == (10, 4)
@@ -380,3 +380,24 @@ def test_layers(self):
 
         avg = b.random(size=10000).mean(axis=0)
         npt.assert_array_almost_equal(avg, 0.5 * np.ones_like(b), decimal=2)
+
+    def test_transformed(self):
+        n = 18
+        at_bats = 45 * np.ones(n, dtype=int)
+        hits = np.random.randint(1, 40, size=n, dtype=int)
+        draws = 50
+
+        with pm.Model() as model:
+            phi = pm.Beta('phi', alpha=1., beta=1.)
+
+            kappa_log = pm.Exponential('logkappa', lam=5.)
+            kappa = pm.Deterministic('kappa', tt.exp(kappa_log))
+
+            thetas = pm.Beta('thetas', alpha=phi*kappa, beta=(1.0-phi)*kappa, shape=n)
+
+            y = pm.Binomial('y', n=at_bats, p=thetas, shape=n, observed=hits)
+            gen = pm.sample_prior_predictive(draws)
+
+        assert gen['phi'].shape == (draws,)
+        assert gen['y'].shape == (draws, n)
+        assert 'thetas_logodds__' in gen