BUG: pivot/unstack leading to too many items should raise exception (#23512)

Author: sweb

doc/source/whatsnew/v0.24.0.rst

@@ -1649,6 +1649,7 @@ Reshaping
 - :meth:`DataFrame.nlargest` and :meth:`DataFrame.nsmallest` now returns the correct n values when keep != 'all' also when tied on the first columns (:issue:`22752`)
 - Constructing a DataFrame with an index argument that wasn't already an instance of :class:`~pandas.core.Index` was broken (:issue:`22227`).
 - Bug in :class:`DataFrame` prevented list subclasses to be used to construction (:issue:`21226`)
+- Bug in :func:`DataFrame.unstack` and :func:`DataFrame.pivot_table` returning a missleading error message when the resulting DataFrame has more elements than int32 can handle. Now, the error message is improved, pointing towards the actual problem (:issue:`20601`)
 
 .. _whatsnew_0240.bug_fixes.sparse:
 

pandas/core/reshape/pivot.py

@@ -78,8 +78,6 @@ def pivot_table(data, values=None, index=None, columns=None, aggfunc='mean',
                 pass
         values = list(values)
 
-    # group by the cartesian product of the grouper
-    # if we have a categorical
     grouped = data.groupby(keys, observed=False)
     agged = grouped.agg(aggfunc)
     if dropna and isinstance(agged, ABCDataFrame) and len(agged.columns):

pandas/core/reshape/reshape.py

@@ -109,6 +109,21 @@ def __init__(self, values, index, level=-1, value_columns=None,
         self.removed_level = self.new_index_levels.pop(self.level)
         self.removed_level_full = index.levels[self.level]
 
+        # Bug fix GH 20601
+        # If the data frame is too big, the number of unique index combination
+        # will cause int32 overflow on windows environments.
+        # We want to check and raise an error before this happens
+        num_rows = np.max([index_level.size for index_level
+                           in self.new_index_levels])
+        num_columns = self.removed_level.size
+
+        # GH20601: This forces an overflow if the number of cells is too high.
+        num_cells = np.multiply(num_rows, num_columns, dtype=np.int32)
+
+        if num_rows > 0 and num_columns > 0 and num_cells <= 0:
+            raise ValueError('Unstacked DataFrame is too big, '
+                             'causing int32 overflow')
+
         self._make_sorted_values_labels()
         self._make_selectors()
 

pandas/tests/reshape/test_pivot.py

@@ -1272,6 +1272,17 @@ def test_pivot_string_func_vs_func(self, f, f_numpy):
                                aggfunc=f_numpy)
         tm.assert_frame_equal(result, expected)
 
+    @pytest.mark.slow
+    def test_pivot_number_of_levels_larger_than_int32(self):
+        # GH 20601
+        df = DataFrame({'ind1': np.arange(2 ** 16),
+                        'ind2': np.arange(2 ** 16),
+                        'count': 0})
+
+        with pytest.raises(ValueError, match='int32 overflow'):
+            df.pivot_table(index='ind1', columns='ind2',
+                           values='count', aggfunc='count')
+
 
 class TestCrosstab(object):
 

pandas/tests/test_multilevel.py

@@ -3,6 +3,7 @@
 from warnings import catch_warnings, simplefilter
 import datetime
 import itertools
+
 import pytest
 import pytz
 
@@ -720,6 +721,14 @@ def test_unstack_unobserved_keys(self):
         recons = result.stack()
         tm.assert_frame_equal(recons, df)
 
+    @pytest.mark.slow
+    def test_unstack_number_of_levels_larger_than_int32(self):
+        # GH 20601
+        df = DataFrame(np.random.randn(2 ** 16, 2),
+                       index=[np.arange(2 ** 16), np.arange(2 ** 16)])
+        with pytest.raises(ValueError, match='int32 overflow'):
+            df.unstack()
+
     def test_stack_order_with_unsorted_levels(self):
         # GH 16323