Fix dataset tests

doc/whats-new.rst

@@ -21,6 +21,20 @@ v0.9.0 (unreleased)
 Breaking changes
 ~~~~~~~~~~~~~~~~
 
+- Index coordinates for each dimensions are now optional, and no longer created
+  by default. This has a number of implications:
+
+  - :py:func:`~align` and :py:meth:`~Dataset.reindex` can now error, if
+    dimensions labels are missing and dimensions have different sizes.
+  - Because pandas does not support missing indexes, methods such as
+    ``to_dataframe``/``from_dataframe`` and ``stack``/``unstack`` no longer
+    roundtrip faithfully on all inputs. Use :py:meth:`~Dataset.reset_index` to
+    remove undesired indexes.
+  - ``Dataset.__delitem__`` no longer deletes all variables matching
+    dimension names.
+  - ``DataArray.coords.__delitem__`` is now allowed on variables matching
+    dimension names.
+
 - The default behavior of ``merge`` is now ``compat='no_conflicts'``, so some
   merges will now succeed in cases that previously raised
   ``xarray.MergeError``. Set ``compat='broadcast_equals'`` to restore the

xarray/core/alignment.py

@@ -197,11 +197,11 @@ def reindex_variables(variables, dims, indexes, indexers, method=None,
     """
     # build up indexers for assignment along each dimension
     to_indexers = {}
-    to_shape = {}
     from_indexers = {}
+    # size of reindexed dimensions
+    new_sizes = {}
 
     for name, index in iteritems(indexes):
-        to_shape[name] = index.size
         if name in indexers:
             target = utils.safe_cast_to_index(indexers[name])
             if not index.is_unique:
@@ -210,7 +210,7 @@ def reindex_variables(variables, dims, indexes, indexers, method=None,
                     'index has duplicate values' % name)
             indexer = get_indexer(index, target, method, tolerance)
 
-            to_shape[name] = len(target)
+            new_sizes[name] = len(target)
             # Note pandas uses negative values from get_indexer to signify
             # values that are missing in the index
             # The non-negative values thus indicate the non-missing values
@@ -246,12 +246,17 @@ def var_indexers(var, indexers):
 
     # create variables for the new dataset
     reindexed = OrderedDict()
-    for name, var in iteritems(variables):
-        if name in indexers:
-            # no need to copy, because index data is immutable
-            new_var = IndexVariable(var.dims, indexers[name], var.attrs,
-                                    var.encoding)
+
+    for dim, indexer in indexers.items():
+        if dim in variables:
+            var = variables[dim]
+            args = (var.attrs, var.encoding)
         else:
+            args = ()
+        reindexed[dim] = IndexVariable((dim,), indexers[dim], *args)
+
+    for name, var in iteritems(variables):
+        if name not in indexers:
             assign_to = var_indexers(var, to_indexers)
             assign_from = var_indexers(var, from_indexers)
 
@@ -261,7 +266,8 @@ def var_indexers(var, indexers):
                 dtype, fill_value = _maybe_promote(var.dtype)
 
                 if isinstance(data, np.ndarray):
-                    shape = tuple(to_shape[dim] for dim in var.dims)
+                    shape = tuple(new_sizes.get(dim, size)
+                                  for dim, size in zip(var.dims, var.shape))
                     new_data = np.empty(shape, dtype=dtype)
                     new_data[...] = fill_value
                     # create a new Variable so we can use orthogonal indexing
@@ -291,7 +297,7 @@ def var_indexers(var, indexers):
                 # we neither created a new ndarray nor used fancy indexing
                 new_var = var.copy(deep=copy)
 
-        reindexed[name] = new_var
+            reindexed[name] = new_var
     return reindexed
 
 

xarray/core/coordinates.py

@@ -51,7 +51,7 @@ def to_index(self, ordered_dims=None):
         """
         if ordered_dims is None:
             ordered_dims = self.dims
-        indexes = [self.variables[k].to_index() for k in ordered_dims]
+        indexes = [self.indexes.get(k) for k in ordered_dims]
         return pd.MultiIndex.from_product(indexes, names=list(ordered_dims))
 
     def update(self, other):
@@ -213,9 +213,6 @@ def to_dataset(self):
         return self._to_dataset()
 
     def __delitem__(self, key):
-        if key in self.dims:
-            raise ValueError('cannot delete a coordinate corresponding to a '
-                             'DataArray dimension')
         del self._data._coords[key]
 
 
@@ -244,11 +241,11 @@ def __init__(self, variables, dims):
 
         Arguments
         ---------
-        variables : OrderedDict
+        variables : OrderedDict[Any, Variable]
             Reference to OrderedDict holding variable objects. Should be the
             same dictionary used by the source object.
-        dims : sequence or mapping
-            Should be the same dimensions used by the source object.
+        dims : OrderedDict[Any, int]
+            Map from dimension names to sizes.
         """
         self._variables = variables
         self._dims = dims
@@ -265,10 +262,20 @@ def __contains__(self, key):
         return key in self._dims and key in self._variables
 
     def __getitem__(self, key):
-        if key in self:
-            return self._variables[key].to_index()
-        else:
+        if key not in self._dims:
             raise KeyError(key)
+        return self._variables[key].to_index()
 
     def __unicode__(self):
         return formatting.indexes_repr(self)
+
+    def get(self, key):
+        """Get an index for a dimension, supplying default RangeIndex if needed.
+        """
+        if key not in self._dims:
+            raise KeyError(key)
+
+        if key in self._variables:
+            return self._variables[key].to_index()
+        else:
+            return pd.Index(range(self._dims[key]), name=key)

xarray/core/dataarray.py

@@ -17,7 +17,7 @@
 from .coordinates import (DataArrayCoordinates, LevelCoordinatesSource,
                           Indexes)
 from .dataset import Dataset
-from .pycompat import iteritems, basestring, OrderedDict, zip
+from .pycompat import iteritems, basestring, OrderedDict, zip, range
 from .variable import (as_variable, Variable, as_compatible_data, IndexVariable,
                        default_index_coordinate,
                        assert_unique_multiindex_level_names)
@@ -505,7 +505,7 @@ def encoding(self, value):
     def indexes(self):
         """OrderedDict of pandas.Index objects used for label based indexing
         """
-        return Indexes(self._coords, self.dims)
+        return Indexes(self._coords, OrderedDict(zip(self.dims, self.shape)))
 
     @property
     def coords(self):
@@ -1066,7 +1066,8 @@ def to_pandas(self):
         except KeyError:
             raise ValueError('cannot convert arrays with %s dimensions into '
                              'pandas objects' % self.ndim)
-        return constructor(self.values, *self.indexes.values())
+        indexes = [self.indexes.get(dim) for dim in self.dims]
+        return constructor(self.values, *indexes)
 
     def to_dataframe(self, name=None):
         """Convert this array and its coordinates into a tidy pandas.DataFrame.

xarray/core/dataset.py

@@ -557,22 +557,9 @@ def __setitem__(self, key, value):
 
     def __delitem__(self, key):
         """Remove a variable from this dataset.
-
-        If this variable is a dimension, all variables containing this
-        dimension are also removed.
         """
-        def remove(k):
-            del self._variables[k]
-            self._coord_names.discard(k)
-
-        remove(key)
-
-        if key in self._dims:
-            del self._dims[key]
-            also_delete = [k for k, v in iteritems(self._variables)
-                           if key in v.dims]
-            for key in also_delete:
-                remove(key)
+        del self._variables[key]
+        self._coord_names.discard(key)
 
     # mutable objects should not be hashable
     __hash__ = None
@@ -1224,7 +1211,9 @@ def reindex(self, indexers=None, method=None, tolerance=None, copy=True, **kw_in
         variables = alignment.reindex_variables(
             self.variables, self.dims, self.indexes, indexers, method,
             tolerance, copy=copy)
-        return self._replace_vars_and_dims(variables)
+        coord_names = set(self._coord_names)
+        coord_names.update(indexers)
+        return self._replace_vars_and_dims(variables, coord_names)
 
     def rename(self, name_dict, inplace=False):
         """Returns a new object with renamed variables and dimensions.
@@ -1250,9 +1239,9 @@ def rename(self, name_dict, inplace=False):
         DataArray.rename
         """
         for k, v in name_dict.items():
-            if k not in self:
+            if k not in self and k not in self.dims:
                 raise ValueError("cannot rename %r because it is not a "
-                                 "variable in this dataset" % k)
+                                 "variable or dimension in this dataset" % k)
             if v in self and k != v:
                 raise ValueError('the new name %r already exists' % v)
 
@@ -1339,18 +1328,8 @@ def _stack_once(self, dims, new_dim):
                 else:
                     variables[name] = var.copy(deep=False)
 
-        indexes = self.indexes
-        dim_sizes = self.dims
-
-        levels = []
-        for dim in dims:
-            if dim in indexes:
-                level = indexes[dim]
-            else:
-                level = np.arange(dim_sizes[dim])
-            levels.append(level)
-
         # consider dropping levels that are unused?
+        levels = [self.indexes.get(dim) for dim in dims]
         idx = utils.multiindex_from_product_levels(levels, names=dims)
         variables[new_dim] = IndexVariable(new_dim, idx)
 
@@ -1409,7 +1388,7 @@ def unstack(self, dim):
         if dim not in self.dims:
             raise ValueError('invalid dimension: %s' % dim)
 
-        index = self.indexes[dim]
+        index = self.indexes.get(dim)
         if not isinstance(index, pd.MultiIndex):
             raise ValueError('cannot unstack a dimension that does not have '
                              'a MultiIndex')
@@ -1551,7 +1530,12 @@ def drop(self, labels, dim=None):
         if dim is None:
             return self._drop_vars(labels)
         else:
-            new_index = self.indexes[dim].drop(labels)
+            try:
+                index = self.indexes[dim]
+            except KeyError:
+                raise ValueError(
+                    'dimension %r does not have coordinate labels' % dim)
+            new_index = index.drop(labels)
             return self.loc[{dim: new_index}]
 
     def _drop_vars(self, names):

xarray/test/test_dataarray.py

@@ -651,11 +651,9 @@ def test_coords(self):
         actual = repr(da.coords)
         self.assertEquals(expected, actual)
 
-        with self.assertRaisesRegexp(ValueError, 'cannot delete'):
-            del da['x']
-
-        with self.assertRaisesRegexp(ValueError, 'cannot delete'):
-            del da.coords['x']
+        del da.coords['x']
+        expected = DataArray(da.values, {'y': [0, 1, 2]}, dims=['x', 'y'])
+        self.assertDataArrayIdentical(da, expected)
 
         with self.assertRaisesRegexp(ValueError, 'conflicting MultiIndex'):
             self.mda['level_1'] = np.arange(4)