ENH: option to disable empty string conversion to NaN in file readers

Author: Chang She

pandas/io/parsers.py

@@ -1029,9 +1029,9 @@ def _convert_types(values, na_values):
         return values, na_count
 
     try:
-        result = lib.maybe_convert_numeric(values, na_values)
+        result = lib.maybe_convert_numeric(values, na_values, False)
     except Exception:
-        na_count = lib.sanitize_objects(values, na_values)
+        na_count = lib.sanitize_objects(values, na_values, False)
         result = values
 
     if result.dtype == np.object_:

pandas/io/tests/test_parsers.py

@@ -52,6 +52,32 @@ def setUp(self):
         self.csv2 = os.path.join(self.dirpath, 'test2.csv')
         self.xls1 = os.path.join(self.dirpath, 'test.xls')
 
+    def test_empty_string(self):
+        data = """\
+One,Two,Three
+a,1,one
+b,2,two
+,3,three
+d,4,nan
+e,5,five
+nan,6,
+g,7,seven
+"""
+        df = read_csv(StringIO(data))
+        xp = DataFrame({'One' : ['a', 'b', np.nan, 'd', 'e', np.nan, 'g'],
+                        'Two' : [1,2,3,4,5,6,7],
+                        'Three' : ['one', 'two', 'three', np.nan, 'five',
+                                   np.nan, 'seven']})
+        assert_frame_equal(xp.reindex(columns=df.columns), df)
+
+        df = read_csv(StringIO(data), na_values={'One': [], 'Three': []})
+        xp = DataFrame({'One' : ['a', 'b', '', 'd', 'e', 'nan', 'g'],
+                        'Two' : [1,2,3,4,5,6,7],
+                        'Three' : ['one', 'two', 'three', 'nan', 'five',
+                                   '', 'seven']})
+        assert_frame_equal(xp.reindex(columns=df.columns), df)
+
+
     def test_read_csv(self):
         pass
 

pandas/src/inference.pyx

@@ -241,7 +241,8 @@ def is_date_array(ndarray[object] values):
     return True
 
 
-def maybe_convert_numeric(ndarray[object] values, set na_values):
+def maybe_convert_numeric(ndarray[object] values, set na_values,
+                          convert_empty=True):
     '''
     Type inference function-- convert strings to numeric (potentially) and
     convert to proper dtype array
@@ -275,8 +276,11 @@ def maybe_convert_numeric(ndarray[object] values, set na_values):
             floats[i] = complexes[i] = nan
             seen_float = 1
         elif len(val) == 0:
-            floats[i] = complexes[i] = nan
-            seen_float = 1
+            if convert_empty:
+                floats[i] = complexes[i] = nan
+                seen_float = 1
+            else:
+                raise ValueError('Empty string encountered')
         elif util.is_complex_object(val):
             complexes[i] = val
             seen_complex = 1
@@ -573,7 +577,8 @@ def try_parse_datetime_components(ndarray[object] years, ndarray[object] months,
 
     return result
 
-def sanitize_objects(ndarray[object] values, set na_values):
+def sanitize_objects(ndarray[object] values, set na_values,
+                     convert_empty=True):
     cdef:
         Py_ssize_t i, n
         object val, onan
@@ -585,7 +590,7 @@ def sanitize_objects(ndarray[object] values, set na_values):
 
     for i from 0 <= i < n:
         val = values[i]
-        if val == '' or val in na_values:
+        if (convert_empty and val == '') or (val in na_values):
             values[i] = onan
             na_count += 1
         elif val in memo: