Merge pull request #957 from rizar/no_bias_shallow_energy_computer

Optional bias in ShallowEnergyComputer

blocks/bricks/attention.py

@@ -415,11 +415,20 @@ def get_dim(self, name):
 
 
 class ShallowEnergyComputer(Sequence, Initializable, Feedforward):
-    """A simple energy computer: first tanh, then weighted sum."""
+    """A simple energy computer: first tanh, then weighted sum.
+
+    Parameters
+    ----------
+    use_bias : bool, optional
+        Whether a bias should be added to the energies. Does not change
+        anything if softmax normalization is used to produce the attention
+        weights, but might be useful when e.g. spherical softmax is used.
+
+    """
     @lazy()
-    def __init__(self, **kwargs):
+    def __init__(self, use_bias=False, **kwargs):
         super(ShallowEnergyComputer, self).__init__(
-            [Tanh().apply, Linear(use_bias=False).apply], **kwargs)
+            [Tanh().apply, Linear(use_bias=use_bias).apply], **kwargs)
 
     @property
     def input_dim(self):