Rewrote example of usage of verify_grad, which was incomplete and possibly confusing

doc/extending/unittest.txt

@@ -372,7 +372,8 @@ the given tolerance.
 The parameters are as follows:
 
 * op: something that behaves like an Op instance with a single output
-  (can be a python function combining multiple ops)
+  (can be for instance a python function combining multiple ops, or calling
+  an op with some of the inputs being fixed to specific values).
 
 * pt: the list of numpy.ndarrays to use as inputs to the op
 
@@ -387,11 +388,21 @@ The parameters are as follows:
 Here is an example showing how to use verify_grad:
 
 >>> def test_flatten_outdimNone():
->>>    a = dmatrix()
->>>    # ...
->>>    a_val = theano._asarray([[0,1,2],[3,4,5]], dtype='float64')
->>>    # ...
->>>    tensor.verify_grad(Flatten(), [a_val])
+>>>    # Testing gradient w.r.t. all inputs of an op (in this example the op
+>>>    # being used is Flatten(), which takes a single input).
+>>>    a_val = numpy.asarray([[0,1,2],[3,4,5]], dtype='float64')
+>>>    tensor.verify_grad(tensor.Flatten(), [a_val])
+>>>    # Testing gradient w.r.t. to a subset of an op's inputs. This is useful
+>>>    # in particular when the gradient w.r.t. some of the inputs cannot be
+>>>    # computed by finite difference (e.g. for discrete inputs), which would
+>>>    # cause verify_grad to crash.
+>>>    op = tensor.nnet.crossentropy_softmax_argmax_1hot_with_bias
+>>>    def op_with_fixed_y_idx(x, b):
+>>>        # Note that this op has multiple outputs: we need to pick only one.
+>>>        return op(x, b, y_idx=numpy.asarray([0, 2]))[0]
+>>>    x_val = numpy.asarray([[-1, 0, 1], [3, 2, 1]], dtype='float64')
+>>>    b_val = numpy.asarray([1, 2, 3], dtype='float64')
+>>>    tensor.verify_grad(op_with_fixed_y_idx, [x_val, b_val])
 
 .. note::