# HG changeset patch
# User Joseph Turian <turian@gmail.com>
# Date 1236744562 14400
# Node ID 576eb7f77c35f249278b950fb5a8bb7884aa39dc
# Parent  b282a5c2f76b6d4f15796203aa8ae9d9b9bdcdd6
Trying to fix gradient in logfactorial

diff -r b282a5c2f76b -r 576eb7f77c35 pylearn/algorithms/sandbox/cost.py
--- a/pylearn/algorithms/sandbox/cost.py	Tue Mar 10 20:58:16 2009 -0400
+++ b/pylearn/algorithms/sandbox/cost.py	Wed Mar 11 00:09:22 2009 -0400
@@ -9,6 +9,18 @@
 from theano import tensor, scalar
 import numpy
 
+class UndefinedGradient(Exception):
+    """
+    Raised by UndefinedGradientOp to indicate that the gradient is undefined mathematically.
+    """
+    pass
+from theano import gof
+class UndefinedGradientOp(gof.Op):
+    def perform(self, x=None):
+        if x is not None: raise UndefinedGradient(x)
+        else: raise UndefinedGradient(x)
+undefined_gradient = UndefinedGradientOp()
+
 class LogFactorial(scalar.UnaryScalarOp):
     """
     Compute log x!.
@@ -28,6 +40,8 @@
         return v
     def impl(self, x):
         return LogFactorial.st_impl(x)
+    def grad(self, (x,), (gz,)):
+        undefined_gradient(self)
 #    def grad(self, (x,), (gz,)):
 #        raise NotImplementedError('gradient not defined over discrete values')
 #        return None
diff -r b282a5c2f76b -r 576eb7f77c35 pylearn/algorithms/sandbox/test_cost.py
--- a/pylearn/algorithms/sandbox/test_cost.py	Tue Mar 10 20:58:16 2009 -0400
+++ b/pylearn/algorithms/sandbox/test_cost.py	Wed Mar 11 00:09:22 2009 -0400
@@ -38,7 +38,16 @@
 #        (goutput) = TT.grad(loss, [target])
         f = T.function([], goutput)
         print f()
-        self.failUnless(f() - 33751.7816277 < 1e-5)
+        self.failUnless(numpy.all(f() - numpy.asarray([206., 559.96605666, 558.96605666, 205., 557.96605666, 204., 30473.11077513, 459.96605666] < 1e-5)))
+
+    def test_gradient_fail(self):
+        target = TT.as_tensor([0, 0, 1, 1, 2, 2, 100, 100])
+        output = TT.as_tensor([0., 1, 1., 0, 1, 0, 5, 1])
+        loss = cost.nlpoisson(target, output)
+        (goutput) = TT.grad(loss, [target])
+        f = T.function([], goutput)
+        print f()
+        self.failUnless(numpy.all(f() - numpy.asarray([206., 559.96605666, 558.96605666, 205., 557.96605666, 204., 30473.11077513, 459.96605666] < 1e-5)))
 
 if __name__ == '__main__':
     unittest.main()