Add testing for more use cases

Author: antonwolfy

dpnp/tests/test_special.py

@@ -1,32 +1,42 @@
 import math
 
 import numpy
-from numpy.testing import assert_allclose
+import pytest
+import scipy
+from numpy.testing import assert_allclose, assert_almost_equal
 
 import dpnp
 
+from .helper import (
+    generate_random_numpy_array,
+    get_all_dtypes,
+    get_complex_dtypes,
+)
 
-def test_erf():
-    a = numpy.linspace(2.0, 3.0, num=10)
-    ia = dpnp.array(a)
 
-    expected = numpy.empty_like(a)
-    for idx, val in enumerate(a):
-        expected[idx] = math.erf(val)
+class TestErf:
 
-    result = dpnp.erf(ia)
+    @pytest.mark.parametrize(
+        "dt", get_all_dtypes(no_none=True, no_float16=False, no_complex=True)
+    )
+    def test_basic(self, dt):
+        a = generate_random_numpy_array((2, 5), dtype=dt)
+        ia = dpnp.array(a)
 
-    assert_allclose(result, expected)
+        result = dpnp.special.erf(ia)
+        expected = scipy.special.erf(a)
+        assert_almost_equal(result, expected)
 
+    def test_nan_inf(self):
+        a = numpy.array([numpy.nan, -numpy.inf, numpy.inf])
+        ia = dpnp.array(a)
 
-def test_erf_fallback():
-    a = numpy.linspace(2.0, 3.0, num=10)
-    dpa = dpnp.linspace(2.0, 3.0, num=10)
+        result = dpnp.special.erf(ia)
+        expected = scipy.special.erf(a)
+        assert_allclose(result, expected)
 
-    expected = numpy.empty_like(a)
-    for idx, val in enumerate(a):
-        expected[idx] = math.erf(val)
-
-    result = dpnp.erf(dpa)
-
-    assert_allclose(result, expected)
+    @pytest.mark.parametrize("dt", get_complex_dtypes())
+    def test_complex(self, dt):
+        x = dpnp.empty(5, dtype=dt)
+        with pytest.raises(ValueError):
+            dpnp.special.erf(x)