Merge pull request #4602 from cancan101/number_parsing

Fixes issue with TextFileReader using python engine and thousands != ","

Author: jreback

doc/source/release.rst

@@ -429,7 +429,9 @@ Bug Fixes
     ``ascending`` was being interpreted as ``True`` (:issue:`4839`,
     :issue:`4846`)
   - Fixed ``Panel.tshift`` not working. Added `freq` support to ``Panel.shift`` (:issue:`4853`)
-
+  - Fix an issue in TextFileReader w/ Python engine (i.e. PythonParser) 
+    with thousands != "," (:issue:`4596`)
+    
 pandas 0.12.0
 -------------
 

pandas/io/parsers.py

@@ -1538,7 +1538,7 @@ def _check_thousands(self, lines):
                         nonnum.search(x.strip())):
                     rl.append(x)
                 else:
-                    rl.append(x.replace(',', ''))
+                    rl.append(x.replace(self.thousands, ''))
             ret.append(rl)
         return ret
 

pandas/io/tests/test_cparser.py

@@ -19,7 +19,7 @@
 from pandas import DataFrame, Series, Index, isnull, MultiIndex
 import pandas.io.parsers as parsers
 from pandas.io.parsers import (read_csv, read_table, read_fwf,
-                               TextParser)
+                               TextParser, TextFileReader)
 from pandas.util.testing import (assert_almost_equal, assert_frame_equal,
                                  assert_series_equal, network)
 import pandas.lib as lib
@@ -132,6 +132,16 @@ def test_integer_thousands(self):
 
         expected = [123456, 12500]
         tm.assert_almost_equal(result[0], expected)
+        
+    def test_integer_thousands_alt(self):
+        data = '123.456\n12.500'
+
+        reader = TextFileReader(StringIO(data), delimiter=':',
+                            thousands='.', header=None)
+        result = reader.read()
+
+        expected = [123456, 12500]
+        tm.assert_almost_equal(result[0], expected)
 
     def test_skip_bad_lines(self):
         # too many lines, see #2430 for why