Merge pull request #1229 from taku-y/sample_vp

Hide transformed vars in sample_vp().

Author: twiecki

pymc3/tests/test_advi.py

@@ -1,4 +1,5 @@
 import numpy as np
+import pymc3 as pm
 from pymc3 import Model, Normal, DiscreteUniform, Poisson, switch, Exponential
 from pymc3.theanof import inputvars
 from pymc3.variational.advi import variational_gradient_estimate, advi, advi_minibatch, sample_vp
@@ -202,3 +203,22 @@ def create_minibatch(data):
 
     np.testing.assert_allclose(np.mean(trace['mu']), mu_post, rtol=0.4)
     np.testing.assert_allclose(np.std(trace['mu']), np.sqrt(1. / d), rtol=0.4)
+
+def test_sample_vp():
+    n_samples = 100
+
+    rng = np.random.RandomState(0)
+    xs = rng.binomial(n=1, p=0.2, size=n_samples)
+
+    with pm.Model() as model:
+        p = pm.Beta('p', alpha=1, beta=1)
+        pm.Binomial('xs', n=1, p=p, observed=xs)
+        v_params = advi(n=1000)
+
+    with model:
+        trace = sample_vp(v_params, hide_transformed=True)
+    assert(set(trace.varnames) == set('p'))
+
+    with model:
+        trace = sample_vp(v_params, hide_transformed=False)
+    assert(set(trace.varnames) == set(('p', 'p_logodds_')))

pymc3/variational/advi.py

@@ -325,7 +325,8 @@ def adagrad(grad, param, learning_rate, epsilon, n):
                               tt.sqrt(accu_sum + epsilon))
     return updates
 
-def sample_vp(vparams, draws=1000, model=None, random_seed=20090425):
+def sample_vp(vparams, draws=1000, model=None, random_seed=20090425, 
+              hide_transformed=True):
     """Draw samples from variational posterior.
 
     Parameters
@@ -338,6 +339,8 @@ def sample_vp(vparams, draws=1000, model=None, random_seed=20090425):
         Probabilistic model.
     random_seed : int
         Seed of random number generator.
+    hide_transformed : bool
+        If False, transformed variables are also sampled. Default is True. 
 
     Returns
     -------
@@ -366,8 +369,13 @@ def sample_vp(vparams, draws=1000, model=None, random_seed=20090425):
     samples = theano.clone(vars, updates)
     f = theano.function([], samples)
 
+    # Random variables which will be sampled
+    vars_sampled = [v for v in model.unobserved_RVs if not str(v).endswith('_')] \
+                   if hide_transformed else \
+                   [v for v in model.unobserved_RVs]
+
     varnames = [str(var) for var in model.unobserved_RVs]
-    trace = NDArray(model=model, vars=model.unobserved_RVs)
+    trace = NDArray(model=model, vars=vars_sampled)
     trace.setup(draws=draws, chain=0)
 
     for i in range(draws):