Merge pull request #9127 from shoyer/fix-slow-loc

PERF: fix slow s.loc[[0]]

Author: shoyer

doc/source/whatsnew/v0.16.0.txt

@@ -45,6 +45,7 @@ Performance
 
 .. _whatsnew_0160.performance:
 
+- Fixed a severe performance regression for ``.loc`` indexing with an array or list (:issue:9126:).
 
 Bug Fixes
 ~~~~~~~~~

pandas/core/indexing.py

@@ -1275,9 +1275,8 @@ def _has_valid_type(self, key, axis):
             if isinstance(key, tuple) and isinstance(ax, MultiIndex):
                 return True
 
-            # require at least 1 element in the index
-            idx = _ensure_index(key)
-            if len(idx) and not idx.isin(ax).any():
+            # TODO: don't check the entire key unless necessary
+            if len(key) and np.all(ax.get_indexer_for(key) < 0):
 
                 raise KeyError("None of [%s] are in the [%s]" %
                                (key, self.obj._get_axis_name(axis)))

vb_suite/indexing.py

@@ -209,3 +209,30 @@
 
 frame_iloc_big = Benchmark('df.iloc[:100,0]', setup,
                             start_date=datetime(2013, 1, 1))
+
+#----------------------------------------------------------------------
+# basic tests for [], .loc[], .iloc[] and .ix[]
+
+setup = common_setup + """
+s = Series(np.random.rand(1000000))
+"""
+
+series_getitem_scalar = Benchmark("s[800000]", setup)
+series_getitem_slice = Benchmark("s[:800000]", setup)
+series_getitem_list_like = Benchmark("s[[800000]]", setup)
+series_getitem_array = Benchmark("s[np.arange(10000)]", setup)
+
+series_loc_scalar = Benchmark("s.loc[800000]", setup)
+series_loc_slice = Benchmark("s.loc[:800000]", setup)
+series_loc_list_like = Benchmark("s.loc[[800000]]", setup)
+series_loc_array = Benchmark("s.loc[np.arange(10000)]", setup)
+
+series_iloc_scalar = Benchmark("s.loc[800000]", setup)
+series_iloc_slice = Benchmark("s.loc[:800000]", setup)
+series_iloc_list_like = Benchmark("s.loc[[800000]]", setup)
+series_iloc_array = Benchmark("s.loc[np.arange(10000)]", setup)
+
+series_ix_scalar = Benchmark("s.ix[800000]", setup)
+series_ix_slice = Benchmark("s.ix[:800000]", setup)
+series_ix_list_like = Benchmark("s.ix[[800000]]", setup)
+series_ix_array = Benchmark("s.ix[np.arange(10000)]", setup)