Merge pull request #3456 from nouiz/nanguard

doc/extending/op.txt

@@ -533,6 +533,8 @@ These are the function required to work with gradient.grad().
    the outputs) back to their corresponding shapes and return them as the 
    output of the :func:`R_op` method.
 
+   :ref:`List of op with r op support <R_op_list>`.
+
 Defining an Op: ``mul``
 =======================
 

doc/library/gradient.txt

@@ -20,5 +20,59 @@ function does the underlying work, and is more flexible, but is also more
 awkward to use when :func:`gradient.grad` can do the job.
 
 
+Gradient related functions
+==========================
+
 .. automodule:: theano.gradient
     :members:
+
+.. _R_op_list:
+
+
+List of Implemented R op
+========================
+
+
+See the :ref:`gradient tutorial <tutcomputinggrads>` for the R op documentation.
+
+list of ops that support R-op:
+ * with test [Most is tensor/tests/test_rop.py]
+    * SpecifyShape
+    * MaxAndArgmax
+    * Subtensor
+    * IncSubtensor set_subtensor too
+    * Alloc
+    * Dot
+    * Elemwise
+    * Sum
+    * Softmax
+    * Shape
+    * Join
+    * Rebroadcast
+    * Reshape
+    * Flatten
+    * DimShuffle
+    * Scan [In scan_module/tests/test_scan.test_rop]
+
+ * without test
+    * Split
+    * ARange
+    * ScalarFromTensor
+    * AdvancedSubtensor1
+    * AdvancedIncSubtensor1
+    * AdvancedIncSubtensor
+
+Partial list of ops without support for R-op:
+
+ * All sparse ops
+ * All linear algebra ops.
+ * PermuteRowElements
+ * Tile
+ * AdvancedSubtensor
+ * TensorDot
+ * Outer
+ * Prod
+ * MulwithoutZeros
+ * ProdWithoutZeros
+ * CAReduce(for max,... done for MaxAndArgmax op)
+ * MaxAndArgmax(only for matrix on axis 0 or 1)

doc/library/tensor/basic.txt

@@ -1746,55 +1746,3 @@ Gradient / Differentiation
 
 See the :ref:`gradient <libdoc_gradient>` page for complete documentation
 of the gradient module.
-
-.. _R_op_list:
-
-
-List of Implemented R op
-========================
-
-
-See the :ref:`gradient tutorial <tutcomputinggrads>` for the R op documentation.
-
-list of ops that support R-op:
- * with test [Most is tensor/tests/test_rop.py]
-    * SpecifyShape
-    * MaxAndArgmax
-    * Subtensor
-    * IncSubtensor set_subtensor too
-    * Alloc
-    * Dot
-    * Elemwise
-    * Sum
-    * Softmax
-    * Shape
-    * Join
-    * Rebroadcast
-    * Reshape
-    * Flatten
-    * DimShuffle
-    * Scan [In scan_module/tests/test_scan.test_rop]
-
- * without test
-    * Split
-    * ARange
-    * ScalarFromTensor
-    * AdvancedSubtensor1
-    * AdvancedIncSubtensor1
-    * AdvancedIncSubtensor
-
-Partial list of ops without support for R-op:
-
- * All sparse ops
- * All linear algebra ops.
- * PermuteRowElements
- * Tile
- * AdvancedSubtensor
- * TensorDot
- * Outer
- * Prod
- * MulwithoutZeros
- * ProdWithoutZeros
- * CAReduce(for max,... done for MaxAndArgmax op)
- * MaxAndArgmax(only for matrix on axis 0 or 1)
-

doc/tutorial/gradients.txt

@@ -238,6 +238,7 @@ array([[ 0.,  0.],
     as the input parameter, while the result of the *R-operator* has a shape similar
     to that of the output.
 
+   :ref:`List of op with r op support <R_op_list>`.
 
 Hessian times a Vector
 ======================

theano/compile/nanguardmode.py

@@ -263,14 +263,14 @@ class NanGuardMode(Mode):
                     error = True
             if big_is_error:
                 err = False
-                if var.size == 0:
+                if isinstance(var, theano.gof.type.CDataType._cdata_type):
                     err = False
                 elif cuda.cuda_available and isinstance(var, cuda.CudaNdarray):
                     err = (f_gpuabsmax(var.reshape(var.size)) > 1e10)
-                elif isinstance(var, theano.gof.type.CDataType._cdata_type):
-                    err = False
                 elif isinstance(var, np.random.mtrand.RandomState):
                     err = False
+                elif var.size == 0:
+                    err = False
                 else:
                     err = (np.abs(var).max() > 1e10)
                 if err: