Merge pull request #3189 from nouiz/fix_nan

theano/sandbox/gpuarray/tests/test_basic_ops.py

@@ -384,6 +384,14 @@ class G_reshape(test_basic.T_reshape):
         assert self.op == GpuReshape
 
 
+class G_comparison(test_basic.test_comparison):
+    def setUp(self):
+        utt.seed_rng()
+        self.mode = mode_with_gpu
+        self.shared = gpuarray_shared_constructor
+        self.dtypes = ['float64', 'float32']
+
+
 class G_Join_and_Split(test_basic.T_Join_and_Split):
     def setUp(self):
         super(G_Join_and_Split, self).setUp()

theano/sandbox/gpuarray/type.py

@@ -121,7 +121,20 @@ class GpuArrayType(Type):
             return False
         if a.typecode != b.typecode:
             return False
-        return numpy.asarray(compare(a, '==', b)).all()
+        a_eq_b = numpy.asarray(compare(a, '==', b))
+        if a_eq_b.all():
+            return True
+
+        # maybe the trouble is that there are NaNs
+        a = numpy.asarray(a)
+        b = numpy.asarray(b)
+
+        a_missing = numpy.isnan(a)
+        if a_missing.any():
+            b_missing = numpy.isnan(b)
+            return numpy.all(a_eq_b + (a_missing == b_missing))
+        else:
+            return False
 
     @staticmethod
     def values_eq_approx(a, b,
@@ -157,7 +170,15 @@ class GpuArrayType(Type):
                             op_tmpl="res[i] = (fabs(%%(a)s - %%(b)s) <"
                             "(%(atol_)s + %(rtol_)s * fabs(%%(b)s)))" %
                             locals())
-            return numpy.asarray(res).all()
+            ret = numpy.asarray(res).all()
+            if ret:
+                return True
+            # maybe the trouble is that there are NaNs
+            an = numpy.asarray(a)
+            bn = numpy.asarray(b)
+            return tensor.TensorType.values_eq_approx(
+                an, bn, allow_remove_inf=allow_remove_inf,
+                allow_remove_nan=allow_remove_nan, rtol=rtol, atol=atol)
 
     @staticmethod
     def may_share_memory(a, b):

theano/scalar/basic.py

@@ -1096,8 +1096,6 @@ class EQ(LogicalComparison):
     def c_code(self, node, name, inputs, outputs, sub):
         (x, y) = inputs
         (z,) = outputs
-        if node.inputs[0].type in complex_types:
-            raise NotImplementedError()
         return "%(z)s = (%(x)s == %(y)s);" % locals()
 eq = EQ()
 
@@ -2104,7 +2102,7 @@ class Sgn(UnaryScalarOp):
         (z,) = outputs
         type = node.inputs[0].type
         if type in float_types:
-            return "%(z)s = (%(x)s >= 0) ? (%(x)s == 0) ? 0.0 : 1.0 : -1.0;" % locals()
+            return '%(z)s = (%(x)s > 0) ? 1. : ((%(x)s < 0) ? -1. : (isnan(%(x)s) ? NAN : 0.));' % locals()
         if type in int_types:
             return "%(z)s = (%(x)s >= 0) ? (%(x)s == 0) ? 0 : 1 : -1;" % locals()
         raise TypeError()  # complex has no sgn
@@ -2112,7 +2110,7 @@ class Sgn(UnaryScalarOp):
     def c_code_cache_version(self):
         s = super(Sgn, self).c_code_cache_version()
         if s:
-            return (3,) + s
+            return (4,) + s
         else:  # if parent is unversioned, we are too
             return s
 sgn = Sgn(same_out_nocomplex, name='sgn')