Mercurial > pylearn

changeset 723:2881c67026c1

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
* when creating the sgd minimizer, the user can ask that step_cost compute other outputs at the same same. This allows for example to compute the cost, at the same time as the output class (more efficient)
author desjagui@atchoum.iro.umontreal.ca
date Tue, 26 May 2009 18:45:09 -0400
parents e915f5c9bb21
children 98a99aafd14a
files pylearn/algorithms/sgd.py
diffstat 1 files changed, 11 insertions(+), 6 deletions(-) [+]
line wrap: on
line diff
--- a/pylearn/algorithms/sgd.py	Tue May 26 17:39:43 2009 -0400
+++ b/pylearn/algorithms/sgd.py	Tue May 26 18:45:09 2009 -0400
@@ -5,7 +5,9 @@
 
 class StochasticGradientDescent(theano.Module):
     """Fixed stepsize gradient descent"""
-    def __init__(self, args, cost, params, gradients=None, stepsize=None, updates=None):
+    def __init__(self, args, cost, params, 
+                 gradients=None, stepsize=None, 
+                 updates=None, auxout=None):
         """
         :param stepsize: the step to take in (negative) gradient direction
         :type stepsize: None, scalar value, or scalar TensorVariable
@@ -13,6 +15,8 @@
         :param updates: extra symbolic updates to make when evating either step or step_cost
         (these override the gradients if necessary)
         :type updatess: dict Variable -> Variable
+        :type auxout: auxiliary outputs, list containing output symbols to 
+                      compute at the same time as cost (for efficiency)
         """
         super(StochasticGradientDescent, self).__init__()
         self.stepsize_init = None
@@ -34,12 +38,12 @@
         if updates is not None:
             self._updates.update(updates)
 
-
+        auxout = auxout if auxout else []
         self.step = theano.Method(
-                args, [],
+                args, auxout,
                 updates=self._updates)
         self.step_cost = theano.Method(
-                args, cost,
+                args, [cost]+auxout,
                 updates=self._updates)
 
     updates = property(lambda self: self._updates.copy())
@@ -52,6 +56,7 @@
     
     :returns: standard minimizer constructor f(args, cost, params, gradient=None)
     """
-    def f(args, cost, params, gradient=None, updates=None):
-        return StochasticGradientDescent(args, cost, params, gradient, stepsize, updates=updates)
+    def f(args, cost, params, gradient=None, updates=None, auxout=None):
+        return StochasticGradientDescent(args, cost, params, gradient, stepsize,
+                updates=updates, auxout=auxout)
     return f