Do not test div on denominator too close to 0.

theano/tensor/tests/test_basic.py

@@ -345,6 +345,17 @@ def rand(*shape):
     return r * 2 - 1
 
 
+def rand_nonzero(shape, eps=3e-4):
+    """Like rand, but the absolute value has to be at least eps"""
+    # covers [0, 1)
+    r = numpy.asarray(numpy.random.rand(*shape), dtype=config.floatX)
+    # covers [0, (1 - eps) / 2) U [(1 + eps) / 2, 1)
+    r = r * (1 - eps) + eps * (r >= 0.5)
+    # covers [-1, -eps) U [eps, 1)
+    r = r * 2 - 1
+    return r
+
+
 def randint(*shape):
     return numpy.random.random_integers(-5, 5, shape)
 
@@ -355,6 +366,10 @@ def randcomplex(*shape):
     return numpy.complex128(2 * r - 1)
 
 
+def randcomplex_nonzero(shape, eps=1e-4):
+    return numpy.complex128(rand_nonzero(shape, eps))
+
+
 def randint_nonzero(*shape):
     r = numpy.random.random_integers(-5, 4, shape)
     return r + (r == 0) * 5
@@ -559,6 +574,7 @@ MulInplaceTester = makeBroadcastTester(op = inplace.mul_inplace,
                                          grad = _grad_broadcast_binary_normal,
                                          inplace = True)
 
+
 def copymod(dct, without=[], **kwargs):
     """Return dct but with the keys named by args removed, and with
     kwargs added.
@@ -572,12 +588,12 @@ def copymod(dct, without=[], **kwargs):
     return rval
 
 _good_broadcast_div_mod_normal_float_no_complex = dict(
-    same_shapes=(rand(2, 3), rand(2, 3)),
-    scalar=(rand(2, 3), rand(1, 1)),
-    row=(rand(2, 3), rand(1, 3)),
-    column=(rand(2, 3), rand(2, 1)),
+    same_shapes=(rand(2, 3), rand_nonzero((2, 3))),
+    scalar=(rand(2, 3), rand_nonzero((1, 1))),
+    row=(rand(2, 3), rand_nonzero((1, 3))),
+    column=(rand(2, 3), rand_nonzero((2, 1))),
     dtype_mixup_1=(rand(2, 3), randint_nonzero(2, 3)),
-    dtype_mixup_2=(randint_nonzero(2, 3), rand(2, 3)),
+    dtype_mixup_2=(randint_nonzero(2, 3), rand_nonzero((2, 3))),
     integer=(randint(2, 3), randint_nonzero(2, 3)),
     uinteger=(randint(2, 3).astype("uint8"),
               randint_nonzero(2, 3).astype("uint8")),
@@ -588,8 +604,8 @@ _good_broadcast_div_mod_normal_float_no_complex = dict(
 _good_broadcast_div_mod_normal_float_inplace = copymod(
     _good_broadcast_div_mod_normal_float_no_complex,
     empty1=(numpy.asarray([]), numpy.asarray([1])),
-    complex1=(randcomplex(2, 3), randcomplex(2, 3)),
-    complex2=(randcomplex(2, 3), rand(2, 3)),
+    complex1=(randcomplex(2, 3), randcomplex_nonzero((2, 3))),
+    complex2=(randcomplex(2, 3), rand_nonzero((2, 3))),
     # Inplace on the first element. Must have the same type.
     #complex3=(rand(2, 3) ,randcomplex(2, 3)),
     )
@@ -600,18 +616,19 @@ _good_broadcast_div_mod_normal_float = copymod(
     )
 
 
-_grad_broadcast_div_mod_normal = dict(same_shapes = (rand(2, 3), rand(2, 3)),
-                                      scalar = (rand(2, 3), rand(1, 1)),
-                                      row = (rand(2, 3), rand(1, 3)),
-                                      column = (rand(2, 3), rand(2, 1)),
-                                      #complex1 = (randcomplex(2,3),randcomplex(2,3)),
-                                      #complex2 = (randcomplex(2,3),rand(2,3)),
-                                      #complex3 = (rand(2,3),randcomplex(2,3)),
-                                      #dtype_mixup_1 = (rand(2, 3), randint_nonzero(2, 3)),
-                                      #dtype_mixup_2 = (randint_nonzero(2, 3), rand(2, 3)),
-                                      #empty1 = (numpy.asarray([]), numpy.asarray([1.])),
-                                      #empty2 = (numpy.asarray([0]), numpy.asarray([])),
-                                   )
+_grad_broadcast_div_mod_normal = dict(
+        same_shapes=(rand(2, 3), rand_nonzero((2, 3))),
+        scalar=(rand(2, 3), rand_nonzero((1, 1))),
+        row=(rand(2, 3), rand_nonzero((1, 3))),
+        column=(rand(2, 3), rand_nonzero((2, 1))),
+        #complex1=(randcomplex(2, 3), randcomplex_nonzero((2, 3))),
+        #complex2=(randcomplex(2, 3), rand_nonzero((2, 3))),
+        #complex3=(rand(2, 3), randcomplex_nonzero((2, 3))),
+        #dtype_mixup_1=(rand(2, 3), randint_nonzero(2, 3)),
+        #dtype_mixup_2=(randint_nonzero(2, 3), rand_nonzero((2, 3))),
+        #empty1=(numpy.asarray([]), numpy.asarray([1.])),
+        #empty2=(numpy.asarray([0]), numpy.asarray([])),
+        )
 
 div_grad_rtol=None
 if config.floatX=='float32':