Improve random for ObservedRV (#3036)

To generate random array for ObservedRV in sample_ppc and sample_prior_predictive a shape kwarg sometimes is needed to make sure the generated array has the same shape as the observed.
This PR improved that to make sure the shape is inferred during random number generation.
Previously I tried to add the shape attr to ObservedRV at the model creation, but turns out it breaks the sample_ppc for theano.shared observation. I try a different approach here to always reset the distribution shape at generation.

Author: Junpeng Lao

pymc3/distributions/distribution.py

@@ -383,7 +383,17 @@ def _draw_value(param, point=None, givens=None, size=None):
         elif (hasattr(param, 'distribution') and
                 hasattr(param.distribution, 'random') and
                 param.distribution.random is not None):
-            return param.distribution.random(point=point, size=size)
+            # reset the dist shape for ObservedRV
+            if hasattr(param, 'observations'):
+                dist_tmp = param.distribution
+                try:
+                    distshape = param.observations.shape.eval()
+                except AttributeError:
+                    distshape = param.observations.shape
+                dist_tmp.shape = distshape
+                return dist_tmp.random(point=point, size=size)
+            else:
+                return param.distribution.random(point=point, size=size)
         else:
             if givens:
                 variables, values = list(zip(*givens))

pymc3/tests/test_sampling.py

@@ -353,6 +353,7 @@ def test_ignores_observed(self):
         assert (prior['mu'] < 90).all()
         assert (prior['positive_mu'] > 90).all()
         assert (prior['x_obs'] < 90).all()
+        assert prior['x_obs'].shape == (500, 200)
         npt.assert_array_almost_equal(prior['positive_mu'], np.abs(prior['mu']), decimal=4)
 
     def test_respects_shape(self):
@@ -395,9 +396,28 @@ def test_transformed(self):
 
             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)
+            y = pm.Binomial('y', n=at_bats, p=thetas, observed=hits)
             gen = pm.sample_prior_predictive(draws)
 
         assert gen['phi'].shape == (draws,)
         assert gen['y'].shape == (draws, n)
-        assert 'thetas_logodds__' in gen
+        assert 'thetas_logodds__' in gen
+
+    def test_shared(self):
+        n1 = 10
+        obs = shared(np.random.rand(n1) < .5)
+        draws = 50
+
+        with pm.Model() as m:
+            p = pm.Beta('p', 1., 1.)
+            y = pm.Bernoulli('y', p, observed=obs)
+            gen1 = pm.sample_prior_predictive(draws)
+
+        assert gen1['y'].shape == (draws, n1)
+
+        n2 = 20
+        obs.set_value(np.random.rand(n2) < .5)
+        with m:
+            gen2 = pm.sample_prior_predictive(draws)
+
+        assert gen2['y'].shape == (draws, n2)