--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/examples/theano_update.py	Mon Aug 04 16:29:30 2008 -0400
@@ -0,0 +1,56 @@
+import theano
+from theano import tensor
+
+import numpy
+
+# Two scalar symbolic variables
+a = tensor.scalar()
+b = tensor.scalar()
+
+# Definition of output symbolic variable
+c = a * b
+# Definition of the function computing it
+fprop = theano.function([a,b], [c])
+
+# Initialize numerical variables
+a_val = numpy.array(12.)
+b_val = numpy.array(2.)
+print 'a_val =', a_val
+print 'b_val =', b_val
+
+# Numerical value of output is returned by the call to "fprop"
+c_val = fprop(a_val, b_val)
+print 'c_val =', c_val
+
+
+# Definition of simple update (increment by one)
+new_b = b + 1
+update = theano.function([b], [new_b])
+
+# New numerical value of b is returned by the call to "update"
+b_val = update(b_val)
+print 'new b_val =', b_val
+# We can use the new value in "fprop"
+c_val = fprop(a_val, b_val)
+print 'c_val =', c_val
+
+
+# Definition of in-place update (increment by one)
+re_new_b = tensor.add_inplace(b, 1.)
+re_update = theano.function([b], [re_new_b])
+
+# "re_update" can be used the same way as "update"
+b_val = re_update(b_val)
+print 'new b_val =', b_val
+# We can use the new value in "fprop"
+c_val = fprop(a_val, b_val)
+print 'c_val =', c_val
+
+# It is not necessary to keep the return value when the update is done in place
+re_update(b_val)
+print 'new b_val =', b_val
+c_val = fprop(a_val, b_val)
+print 'c_val =', c_val
+
+
+