Make svgd use momentum updates.

pymc-devs · Mar 13, 2017 · 2d67044 · 2d67044
1 parent 827e515
commit 2d67044
Showing 1 changed file with 3 additions and 2 deletions.
diff --git a/pymc3/variational/svgd.py b/pymc3/variational/svgd.py
@@ -7,7 +7,7 @@
 import theano
 import theano.tensor as tt
 from tqdm import tqdm
-from .updates import adagrad
+from .updates import adagrad, apply_momentum
 
 import pymc3 as pm
 from pymc3.model import modelcontext
@@ -91,7 +91,8 @@ def svgd(vars=None, n=5000, n_particles=100, jitter=.01,
     logp_grad_vec = _make_vectorized_logp_grad(vars, model, theta)
     svgd_grad = -1 * _svgd_gradient(vars, model, theta, logp_grad_vec) # maximize
 
-    svgd_updates = optimizer([svgd_grad], [theta], learning_rate=1e-3)
+    svgd_updates = optimizer([svgd_grad], [theta], learning_rate=1.)
+    svgd_updates = apply_momentum(svgd_updates, [theta], momentum=0.9)
 
     i = tt.iscalar('i')
     svgd_step = theano.function([i], [i],