CLN: Remove special handling of nans in the float64-case of isin

asv_bench/benchmarks/series_methods.py

@@ -38,6 +38,27 @@ def time_isin(self, dtypes):
         self.s.isin(self.values)
 
 
+class IsInFloat64(object):
+
+    def setup(self):
+        self.small = Series([1, 2], dtype=np.float64)
+        self.many_different_values = np.arange(10**6, dtype=np.float64)
+        self.few_different_values = np.zeros(10**7, dtype=np.float64)
+        self.only_nans_values = np.full(10**7, np.nan, dtype=np.float64)
+
+    def time_isin_many_different(self):
+        # runtime is dominated by creation of the lookup-table
+        self.small.isin(self.many_different_values)
+
+    def time_isin_few_different(self):
+        # runtime is dominated by creation of the lookup-table
+        self.small.isin(self.few_different_values)
+
+    def time_isin_nan_values(self):
+        # runtime is dominated by creation of the lookup-table
+        self.small.isin(self.few_different_values)
+
+
 class IsInForObjects(object):
 
     def setup(self):

pandas/_libs/hashtable_func_helper.pxi.in

@@ -210,10 +210,10 @@ def duplicated_{{dtype}}({{scalar}}[:] values, object keep='first'):
 @cython.boundscheck(False)
 {{if dtype == 'object'}}
 
-def ismember_{{dtype}}(ndarray[{{scalar}}] arr, ndarray[{{scalar}}] values, bint hasnans=0):
+def ismember_{{dtype}}(ndarray[{{scalar}}] arr, ndarray[{{scalar}}] values):
 {{else}}
 
-def ismember_{{dtype}}({{scalar}}[:] arr, {{scalar}}[:] values, bint hasnans=0):
+def ismember_{{dtype}}({{scalar}}[:] arr, {{scalar}}[:] values):
 {{endif}}
 
     """
@@ -224,7 +224,6 @@ def ismember_{{dtype}}({{scalar}}[:] arr, {{scalar}}[:] values, bint hasnans=0):
     ----------
     arr : {{dtype}} ndarray
     values : {{dtype}} ndarray
-    hasnans : bint, optional
 
     Returns
     -------
@@ -259,19 +258,13 @@ def ismember_{{dtype}}({{scalar}}[:] arr, {{scalar}}[:] values, bint hasnans=0):
     for i in range(n):
         val = arr[i]
         k = kh_get_{{ttype}}(table, <PyObject*> val)
-        if k != table.n_buckets:
-            result[i] = 1
-        else:
-            result[i] = hasnans and val != val
+        result[i] = (k != table.n_buckets)
     {{else}}
     with nogil:
         for i in range(n):
             val = arr[i]
             k = kh_get_{{ttype}}(table, val)
-            if k != table.n_buckets:
-                result[i] = 1
-            else:
-                result[i] = hasnans and val != val
+            result[i] = (k != table.n_buckets)
     {{endif}}
 
     kh_destroy_{{ttype}}(table)

pandas/core/algorithms.py

@@ -437,8 +437,7 @@ def isin(comps, values):
         try:
             values = values.astype('float64', copy=False)
             comps = comps.astype('float64', copy=False)
-            checknull = isna(values).any()
-            f = lambda x, y: htable.ismember_float64(x, y, checknull)
+            f = lambda x, y: htable.ismember_float64(x, y)
         except (TypeError, ValueError):
             values = values.astype(object)
             comps = comps.astype(object)

pandas/tests/test_algos.py

@@ -632,6 +632,28 @@ def test_different_nan_objects(self):
         result = algos.isin(comps, vals)
         tm.assert_numpy_array_equal(expected, result)
 
+    def test_different_nans_as_float64(self):
+        # GH 21866
+        # create different nans from bit-patterns,
+        # these nans will land in different buckets in the hash-table
+        # if no special care is taken
+        NAN1 = struct.unpack("d", struct.pack("=Q", 0x7ff8000000000000))[0]
+        NAN2 = struct.unpack("d", struct.pack("=Q", 0x7ff8000000000001))[0]
+        assert NAN1 != NAN1
+        assert NAN2 != NAN2
+
+        # check that NAN1 and NAN2 are equivalent:
+        arr = np.array([NAN1, NAN2], dtype=np.float64)
+        lookup1 = np.array([NAN1], dtype=np.float64)
+        result = algos.isin(arr, lookup1)
+        expected = np.array([True, True])
+        tm.assert_numpy_array_equal(result, expected)
+
+        lookup2 = np.array([NAN2], dtype=np.float64)
+        result = algos.isin(arr, lookup2)
+        expected = np.array([True, True])
+        tm.assert_numpy_array_equal(result, expected)
+
 
 class TestValueCounts(object):