Merge pull request #2214 from astrofrog/categorical-nd

Allow CategoricalComponent to be n-dimensional

CHANGES.md

@@ -12,6 +12,8 @@ v1.2.0 (unreleased)
 * Expose ``DataCollection.delay_link_manager_update`` which can be used
   to delay any updating to the link tree when adding datasets. [#2225]
 
+* Allow CategoricalComponents to be n-dimensional. [#2214]
+
 v1.1.0 (2021-07-21)
 -------------------
 

glue/core/component.py

@@ -148,7 +148,7 @@ def autotyped(cls, data, units=None):
         if data.dtype.kind == 'M':
             return DateTimeComponent(data)
 
-        n = coerce_numeric(data)
+        n = coerce_numeric(data.ravel()).reshape(data.shape)
 
         thresh = 0.5
         try:
@@ -389,8 +389,8 @@ def __init__(self, categorical_data, categories=None, jitter=None, units=None):
 
         self._data = categorical_ndarray(categorical_data, copy=False, categories=categories)
 
-        if self._data.ndim != 1:
-            raise ValueError("Categorical Data must be 1-dimensional")
+        if self._data.ndim < 1:
+            raise ValueError("Categorical Data must be at least 1-dimensional")
 
         self.jitter(method=jitter)
 
@@ -435,7 +435,7 @@ def jitter(self, method=None):
         Parameters
         ----------
         method : {None, 'uniform'}
-            If `None`, not jittering is done (or any jittering is undone).
+            If `None`, no jittering is done (or any jittering is undone).
             If ``'uniform'``, the codes are randomized by a uniformly
             distributed random variable.
         """

glue/core/subset.py

@@ -684,7 +684,7 @@ def to_mask(self, data, view=None):
         x = data[self.att, view]
         result = self.roi.contains(x, None)
         assert x.shape == result.shape
-        return result.ravel()
+        return result
 
     def copy(self):
         result = CategoricalROISubsetState()

glue/core/tests/test_component.py

@@ -183,6 +183,23 @@ def test_valueerror_on_bad_jitter(self):
             cat_comp.jitter(method='this will never be a jitter method')
 
 
+def test_nd_categorical_component():
+    data = [['a', 'b'], ['c', 'b']]
+    cat_comp = CategoricalComponent(data)
+    np.testing.assert_equal(cat_comp.data, data)
+    np.testing.assert_equal(cat_comp.labels, data)
+    np.testing.assert_equal(cat_comp.codes, [[0, 1], [2, 1]])
+    np.testing.assert_equal(cat_comp.categories, ['a', 'b', 'c'])
+
+
+def test_nd_categorical_component_autotype():
+    x = np.array([['M', 'F'], ['F', 'M']])
+    assert Component.autotyped(x).categorical
+
+    d = Data(x=[['male', 'female', 'male'], ['female', 'female', 'male']])
+    assert d.get_component('x').categorical
+
+
 class TestCoordinateComponent(object):
 
     def setup_method(self, method):

glue/core/tests/test_subset.py

@@ -693,6 +693,18 @@ def test_inherited_properties():
     assert sub.hub is d.hub
 
 
+class TestNDCategoricalSubsets():
+
+    def setup_method(self, method):
+        self.data = Data(sex=[['Male', 'Male', 'Female'], ['Female', 'Male', 'Female']])
+
+    def test_categorical_roi_subset(self):
+        roi = CategoricalROI(['sex', 'Male'])
+        subset = self.data.new_subset()
+        subset.subset_state = CategoricalROISubsetState(att=self.data.id['sex'], roi=roi)
+        np.testing.assert_equal(self.data.subsets[0].to_mask(), [[1, 1, 0], [0, 1, 0]])
+
+
 class TestCloneSubsetStates():
 
     def setup_method(self, method):

glue/utils/array.py

@@ -57,8 +57,8 @@ def unique(array):
     # numpy.unique doesn't handle mixed-types on python3,
     # so we use pandas
     array = np.asarray(array)
-    I, U = pd.factorize(array, sort=True)
-    return U.astype(array.dtype), I
+    I, U = pd.factorize(array.ravel(), sort=True)
+    return U.astype(array.dtype), I.reshape(array.shape)
 
 
 def shape_to_string(shape):

glue/utils/tests/test_array.py

@@ -36,6 +36,14 @@ def test_unique_dtype():
     assert U.dtype.kind in 'SU'
 
 
+def test_unique_nd():
+
+    array = np.array([['a', 'b'], ['c', 'a']])
+    U, I = unique(array)
+    assert_equal(U, ['a', 'b', 'c'])
+    assert_equal(I, [[0, 1], [2, 0]])
+
+
 def test_shape_to_string():
     assert shape_to_string((1, 4, 3)) == "(1, 4, 3)"