allow missing values in columns (#268)

pint_pandas/pint_array.py

@@ -700,7 +700,16 @@ def _from_sequence(cls, scalars, dtype=None, copy=False):
     def _from_sequence_of_strings(cls, scalars, dtype=None, copy=False):
         if not dtype:
             dtype = PintType.construct_from_quantity_string(scalars[0])
-        return cls._from_sequence([dtype.ureg.Quantity(x) for x in scalars], dtype)
+        # cache this lookup as it'll be used for every value
+        _Q = dtype.ureg.Quantity
+
+        def quantity(value):
+            # Pandas seems to pass empty strings as NaNs
+            if isinstance(value, float) and np.isnan(value):
+                return np.nan
+            return _Q(value)
+
+        return cls._from_sequence(list(map(quantity, scalars)), dtype)
 
     @classmethod
     def _from_factorized(cls, values, original):

pint_pandas/testsuite/test_issues.py

@@ -1,4 +1,6 @@
+import io
 import pickle
+from textwrap import dedent
 import time
 
 import numpy as np
@@ -344,3 +346,21 @@ class TestIssue247(BaseExtensionTests):
     result = a / a
     expected = pd.Series([1, 1, 1], dtype="pint[][Float64]")
     tm.assert_series_equal(result, expected)
+
+
+class TestIssue267(BaseExtensionTests):
+    def test_missing_values(self):
+        "make sure that a missing values don't prevent the column from being imported"
+        data = dedent(
+            """\
+            mass
+            0,1lb
+            1,
+            """
+        )
+        df = pd.read_csv(io.StringIO(data), dtype=dict(mass="pint[kg]"))
+        mass = df["mass"]
+        assert mass.dtype == PintType("kg")
+
+        missing = pd.Series([False, True], name="mass")
+        tm.assert_equal(pd.isna(mass), missing)