def test_larger_error(self):
source_assembly = np.random.rand(50, 2)
source_assembly = NeuroidAssembly(
source_assembly,
coords={
'image_id': list(range(source_assembly.shape[0])),
'neuroid_id': list(range(source_assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
source_assembly = source_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
target_assembly = np.random.rand(100, 3)
target_assembly = NeuroidAssembly(
target_assembly,
coords={
'image_id': list(range(target_assembly.shape[0])),
'neuroid_id': list(range(target_assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
target_assembly = target_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
with pytest.raises(Exception):
subset(source_assembly,
target_assembly,
subset_dims=('presentation', ))
def test_smaller_last(self):
source_assembly = np.random.rand(100, 3)
source_assembly = NeuroidAssembly(
source_assembly,
coords={
'image_id': list(range(source_assembly.shape[0])),
'neuroid_id': list(range(source_assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
source_assembly = source_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
target_assembly = source_assembly.sel(
presentation=list(map(lambda x: (50 + x, ), range(50))),
neuroid=list(map(lambda x: (1 + x, ), range(2))))
subset_assembly = subset(source_assembly,
target_assembly,
subset_dims=('presentation', ))
np.testing.assert_array_equal(subset_assembly.coords.keys(),
target_assembly.coords.keys())
for coord_name in target_assembly.coords:
assert all(
subset_assembly[coord_name] == target_assembly[coord_name])
assert (subset_assembly == target_assembly).all()
def test_equal_shifted(self):
target_assembly = np.random.rand(100, 3)
target_assembly = NeuroidAssembly(
target_assembly,
coords={
'image_id': list(range(target_assembly.shape[0])),
'neuroid_id': list(range(target_assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
target_assembly = target_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
shifted_values = np.concatenate(
(target_assembly.values[1:], target_assembly.values[:1]))
shifed_ids = np.array(list(range(shifted_values.shape[0]))) + 1
shifed_ids[-1] = 0
source_assembly = NeuroidAssembly(
shifted_values,
coords={
'image_id': shifed_ids,
'neuroid_id': list(range(shifted_values.shape[1]))
},
dims=['image_id', 'neuroid_id'])
source_assembly = source_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
subset_assembly = subset(source_assembly,
target_assembly,
subset_dims=('presentation', ))
np.testing.assert_array_equal(subset_assembly.coords.keys(),
target_assembly.coords.keys())
assert subset_assembly.shape == target_assembly.shape
def test_repeated_dim_and_adjacent(self):
source_assembly = np.random.rand(5, 5)
source_assembly = NeuroidAssembly(
source_assembly,
coords={
'image_id':
('presentation', list(range(source_assembly.shape[0]))),
'image_meta':
('presentation', np.zeros(source_assembly.shape[0])),
'adjacent': 12
},
dims=['presentation', 'presentation'])
target_assembly = NeuroidAssembly(
np.zeros(3), coords={
'image_id': [0, 2, 3]
}, dims=['image_id']).stack(presentation=('image_id', ))
subset_assembly = subset(source_assembly,
target_assembly,
subset_dims=('presentation', ),
repeat=True)
np.testing.assert_array_equal(subset_assembly.shape, (3, 3))
assert set(subset_assembly['image_id'].values) == set(
target_assembly['image_id'].values)
assert subset_assembly['adjacent'] == 12
def test_repeated_target(self):
source_assembly = np.random.rand(5, 3)
source_assembly = NeuroidAssembly(
source_assembly,
coords={
'image_id': list(range(source_assembly.shape[0])),
'neuroid_id': list(range(source_assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
source_assembly = source_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
target_assembly = NeuroidAssembly(
np.repeat(source_assembly, 2, axis=0),
coords={
'image_id':
np.repeat(list(range(source_assembly.shape[0])), 2, axis=0),
'neuroid_id':
list(range(source_assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
target_assembly = target_assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
subset_assembly = subset(source_assembly,
target_assembly,
subset_dims=('presentation', ),
repeat=True)
np.testing.assert_array_equal(subset_assembly.coords.keys(),
target_assembly.coords.keys())
for coord_name in target_assembly.coords:
assert all(
subset_assembly[coord_name] == target_assembly[coord_name])
np.testing.assert_array_equal(subset_assembly, target_assembly)
assert (subset_assembly == target_assembly).all()
def align(self, source_assembly, target_assembly):
dimensions = list(self._order_dimensions) + list(
set(source_assembly.dims) - set(self._order_dimensions))
source_assembly = source_assembly.transpose(*dimensions)
return subset(source_assembly,
target_assembly,
subset_dims=[self._alignment_dim],
repeat=self._repeat)
def pipe(self, source_assembly, target_assembly):
# check only for equal values, alignment is given by metadata
assert sorted(source_assembly[self._split_coord].values) == sorted(
target_assembly[self._split_coord].values)
if self._split.do_stratify:
assert hasattr(source_assembly, self._stratification_coord)
assert sorted(source_assembly[self._stratification_coord].values) == \
sorted(target_assembly[self._stratification_coord].values)
cross_validation_values, splits = self._split.build_splits(
target_assembly)
split_scores = []
for split_iterator, (train_indices, test_indices), done \
in tqdm(enumerate_done(splits), total=len(splits), desc='cross-validation'):
train_values, test_values = cross_validation_values[
train_indices], cross_validation_values[test_indices]
train_source = subset(source_assembly,
train_values,
dims_must_match=False)
train_target = subset(target_assembly,
train_values,
dims_must_match=False)
assert len(train_source[self._split_coord]) == len(
train_target[self._split_coord])
test_source = subset(source_assembly,
test_values,
dims_must_match=False)
test_target = subset(target_assembly,
test_values,
dims_must_match=False)
assert len(test_source[self._split_coord]) == len(
test_target[self._split_coord])
split_score = yield from self._get_result(train_source,
train_target,
test_source,
test_target,
done=done)
split_score = split_score.expand_dims('split')
split_score['split'] = [split_iterator]
split_scores.append(split_score)
split_scores = Score.merge(*split_scores)
yield split_scores
def pipe(self, assembly):
"""
:param assembly: the assembly to cross-validate over
"""
cross_validation_values, splits = self._split.build_splits(assembly)
split_scores = []
for split_iterator, (train_indices, test_indices), done \
in tqdm(enumerate_done(splits), total=len(splits), desc='cross-validation'):
train_values, test_values = cross_validation_values[
train_indices], cross_validation_values[test_indices]
train = subset(assembly, train_values, dims_must_match=False)
test = subset(assembly, test_values, dims_must_match=False)
split_score = yield from self._get_result(train, test, done=done)
split_score = split_score.expand_dims('split')
split_score['split'] = [split_iterator]
split_scores.append(split_score)
split_scores = Score.merge(*split_scores)
yield split_scores
def test_category_subselection(self):
assembly = get_assembly('dicarlo.MajajHong2015')
categories = np.unique(assembly['category_name'])
target = xr.DataArray([0] * len(categories),
coords={
'category_name': categories
},
dims=['category_name'
]).stack(presentation=['category_name'])
sub_assembly = subset(assembly,
target,
repeat=True,
dims_must_match=False)
assert (assembly == sub_assembly).all()
def test_equal(self):
assembly = np.random.rand(100, 3)
assembly = NeuroidAssembly(assembly,
coords={
'image_id':
list(range(assembly.shape[0])),
'neuroid_id':
list(range(assembly.shape[1]))
},
dims=['image_id', 'neuroid_id'])
assembly = assembly.stack(presentation=('image_id', ),
neuroid=('neuroid_id', ))
subset_assembly = subset(assembly,
assembly,
subset_dims=('presentation', ))
assert (subset_assembly == assembly).all()
