fix a bug in optimization of two consecutive sum. Add a warning when we hit this…

theano/configdefaults.py

@@ -87,3 +87,7 @@ AddConfigVar('warn.argmax_pushdown_bug',
 AddConfigVar('warn.gpusum_01_011_0111_bug',
              "Warn if we are in a case where old version of Theano had a silent bug with GpuSum pattern 01,011 and 0111 when the first dimensions was bigger then 4096. Was fixed 31 may 2010",
              BoolParam(True))
+
+AddConfigVar('warn.sum_sum_bug',
+             "Warn if we are in a case where Theano version between version 9923a40c7b7a and the 2 august 2010(fixed date), generated an error in that case. This happen when their is 2 consecutive sum in the graph, bad code was generated. Was fixed 2 August 2010",
+             BoolParam(True))

theano/tensor/opt.py

@@ -1709,19 +1709,19 @@ def local_sum_sum(node):
                     # do it all at once
                     return [T.Sum(None)(summed.owner.inputs[0])]
 
+                if theano.config.warn.sum_sum_bug:
+                    print "WARNING: Theano version between version 9923a40c7b7a and the w august 2010(fixed date), generated an error in that case. This happen when their is 2 consecutive sum in the graph, bad code was generated. To disable this warning, set the theano flags warn.sum_sum_bug to False."
+                newaxis=list(tuple(summed.owner.op.axis))
                 # figure out which dimensions of the original input are preserved
-                alldims = range(summed.owner.inputs[0].type.ndim)
-                
-                # trim out the dimensions that were removed by the first sum
-                alldims = [d for i,d in enumerate(alldims) if i in summed.owner.op.axis]
-
-                # trim out the dimensions removed by second sum
-                alldims = [d for i,d in enumerate(alldims) if i in node.op.axis]
-
-                # figure out an axis argument that combines the effect of both
-                newaxis = [i for i in xrange(summed.owner.inputs[0].type.ndim)
-                        if i not in alldims]
-
+                for i in node.op.axis:
+                    new_i = i
+                    for ii in summed.owner.op.axis:
+                        if i>=ii:
+                            new_i+=1
+                    assert new_i not in newaxis
+                    newaxis.append(new_i)
+                        
+                assert len(newaxis)==len(list(summed.owner.op.axis)+list(node.op.axis))
                 combined_sum = T.Sum(newaxis)
                 return [combined_sum(summed.owner.inputs[0])]
 

theano/tensor/tests/test_opt.py

@@ -1531,6 +1531,48 @@ def test_constant_get_stabilized():
     #When this error is fixed, the following line should be ok.
     assert f()==800,f()
 
+class T_local_sum_canonicalize(unittest.TestCase):
+    def setUp(self):
+        self.mode = theano.compile.get_default_mode().including('canonicalize')
+
+    def test_local_sum_all_to_none(self):
+        a = T.tensor3()
+        input=numpy.arange(3*3*3).reshape(3,3,3)
+        f = theano.function([a],a.sum())
+        assert len(f.maker.env.nodes)==1
+        assert numpy.allclose(f(input),input.sum())
+
+        f = theano.function([a],a.sum([0,1,2]))
+        assert len(f.maker.env.nodes)==1
+        assert numpy.allclose(f(input),input.sum())
+
+
+        f = theano.function([a],a.sum(0).sum(0).sum(0))
+        assert len(f.maker.env.nodes)==1
+        assert numpy.allclose(f(input),input.sum())
+
+    def test_local_sum_sum(self):
+        a=T.tensor3()
+        input=numpy.arange(3*3*3).reshape(3,3,3)
+        dims=[(0,0),(1,0),(2,0),(0,1),(1,1),(2,1)]
+
+        for d,dd in dims:
+            f = theano.function([a],a.sum(d).sum(dd))
+            assert numpy.allclose(f(input),input.sum(d).sum(dd))
+            assert len(f.maker.env.nodes)==1
+        for d,dd in dims:
+            f = theano.function([a],a.sum(d).sum(dd).sum(0))
+            assert numpy.allclose(f(input),input.sum(d).sum(dd).sum(0))
+            assert len(f.maker.env.nodes)==1
+        for d in [0,1,2]:
+            f = theano.function([a],a.sum(d).sum(None))
+            assert numpy.allclose(f(input),input.sum(d).sum())
+            assert len(f.maker.env.nodes)==1
+        for d in [0,1,2]:
+            f = theano.function([a],a.sum(None).sum())
+            assert numpy.allclose(f(input),input.sum())
+            assert len(f.maker.env.nodes)==1
+            
 class T_local_sum_dimshuffle(unittest.TestCase):
     def setUp(self):
         self.mode = theano.compile.get_default_mode().including('canonicalize')