def test_from_array(self):
x = np.array([
[1, 2, -1, 3],
[-1, -1, np.nan, 1],
[1, 3, 3, 2]
])
expected = np.array([
[0, 1, MV, 2],
[MV, MV, MV, 0],
[0, 2, 2, 1]
])
anno = AnnotationsContainer.from_array(x)
np.testing.assert_equal(anno.annotations, expected)
self.assertEqual(anno.nclasses, 3)
self.assertEqual(anno.labels, [1, 2, 3])
self.assertEqual(anno.missing_values, [np.nan, -1])
# now using a list-of-lists
x = x.tolist()
anno = AnnotationsContainer.from_array(x)
np.testing.assert_equal(anno.annotations, expected)
self.assertEqual(anno.nclasses, 3)
self.assertEqual(anno.labels, [1, 2, 3])
self.assertEqual(anno.missing_values, [np.nan, -1])
# inconsistent number of elements
x[1] = [1, 2]
self.assertRaises(PyannoValueError,
AnnotationsContainer.from_array,
x)
def _new_annotations_fired(self):
"""Create an empty annotations set."""
annotations = CreateNewAnnotationsDialog.create_annotations_dialog()
if annotations is not None:
name = self.application.database.get_available_id()
anno_cont = AnnotationsContainer.from_array(annotations, name=name)
self.set_annotations(anno_cont)
def estimate_quality_instance_level(annotations, pmids):
m, workers = get_M_overall(annotations, pmids)
instance_model = ModelB.create_initial_state(2, len(workers))
anno = AnnotationsContainer.from_array(m, missing_values=[2])
instance_model.map(anno.annotations)
proxy_skill = (instance_model.theta[:,0,0] + instance_model.theta[:,1,1]) / 2.0
return dict(zip(workers, proxy_skill))
def setUp(self):
self.tmp_filename = mktemp(prefix='tmp_pyanno_db_')
# fixtures
self.model1 = ModelA.create_initial_state(4)
self.annotations1 = self.model1.generate_annotations(100)
self.value1 = self.model1.log_likelihood(self.annotations1)
self.anno_container1 = AnnotationsContainer.from_array(
self.annotations1)
self.data_id1 = 'bogus.txt'
self.model2 = ModelB.create_initial_state(4, 8)
self.annotations2 = self.model2.generate_annotations(100)
self.value2 = self.model2.log_likelihood(self.annotations2)
self.anno_container2 = AnnotationsContainer.from_array(
self.annotations2)
self.data_id2 = 'bogus2.txt'
def setUp(self):
self.tmp_filename = mktemp(prefix='tmp_pyanno_db_')
# fixtures
self.model1 = ModelA.create_initial_state(4)
self.annotations1 = self.model1.generate_annotations(100)
self.value1 = self.model1.log_likelihood(self.annotations1)
self.anno_container1 = AnnotationsContainer.from_array(
self.annotations1)
self.data_id1 = 'bogus.txt'
self.model2 = ModelB.create_initial_state(4, 8)
self.annotations2 = self.model2.generate_annotations(100)
self.value2 = self.model2.log_likelihood(self.annotations2)
self.anno_container2 = AnnotationsContainer.from_array(
self.annotations2)
self.data_id2 = 'bogus2.txt'
def _new_annotations_fired(self):
"""Create an empty annotations set."""
annotations = CreateNewAnnotationsDialog.create_annotations_dialog()
if annotations is not None:
name = self.application.database.get_available_id()
anno_cont = AnnotationsContainer.from_array(annotations,
name=name)
self.set_annotations(anno_cont)
def _edit_data_fired(self):
data_view = DataView(data=self.annotations_container.raw_annotations)
data_view.edit_traits(kind='livemodal', parent=self.info.ui.control)
self.annotations_container = AnnotationsContainer.from_array(
data_view.data, name=self.annotations_container.name)
if self.application is not None:
self.application.main_window.set_annotations(
self.annotations_container)
def _edit_data_fired(self):
data_view = DataView(data=self.annotations_container.raw_annotations)
data_view.edit_traits(kind='livemodal', parent=self.info.ui.control)
self.annotations_container = AnnotationsContainer.from_array(
data_view.data,
name = self.annotations_container.name
)
if self.application is not None:
self.application.main_window.set_annotations(
self.annotations_container)
def main():
""" Entry point for standalone testing/debugging. """
from pyanno.modelBt_loopdesign import ModelBtLoopDesign
model = ModelBtLoopDesign.create_initial_state(5)
annotations = model.generate_annotations(2)
anno = AnnotationsContainer.from_array(annotations, name='blah')
model_view = AnnotationsView(annotations_container=anno, model=HasTraits())
model_view.configure_traits()
return model, annotations, model_view
def main():
""" Entry point for standalone testing/debugging. """
from pyanno.modelBt_loopdesign import ModelBtLoopDesign
model = ModelBtLoopDesign.create_initial_state(5)
annotations = model.generate_annotations(2)
anno = AnnotationsContainer.from_array(annotations, name='blah')
model_view = AnnotationsView(annotations_container=anno, model=HasTraits())
model_view.configure_traits()
return model, annotations, model_view
def test_from_file_with_commas(self):
s = """
1,2,3,
-1, -1, 1,
1, 2, 3
"""
buffer = StringIO(s)
anno = AnnotationsContainer._from_file_object(buffer)
expected = np.array([[0, 1, 2], [MV, MV, 0], [0, 1, 2]])
np.testing.assert_equal(anno.annotations, expected)
def test_from_array(self):
x = np.array([[1, 2, -1, 3], [-1, -1, np.nan, 1], [1, 3, 3, 2]])
expected = np.array([[0, 1, MV, 2], [MV, MV, MV, 0], [0, 2, 2, 1]])
anno = AnnotationsContainer.from_array(x)
np.testing.assert_equal(anno.annotations, expected)
self.assertEqual(anno.nclasses, 3)
self.assertEqual(anno.labels, [1, 2, 3])
self.assertEqual(anno.missing_values, [np.nan, -1])
# now using a list-of-lists
x = x.tolist()
anno = AnnotationsContainer.from_array(x)
np.testing.assert_equal(anno.annotations, expected)
self.assertEqual(anno.nclasses, 3)
self.assertEqual(anno.labels, [1, 2, 3])
self.assertEqual(anno.missing_values, [np.nan, -1])
# inconsistent number of elements
x[1] = [1, 2]
self.assertRaises(PyannoValueError, AnnotationsContainer.from_array, x)
def test_from_file_with_commas(self):
s = """
1,2,3,
-1, -1, 1,
1, 2, 3
"""
buffer = StringIO(s)
anno = AnnotationsContainer._from_file_object(buffer)
expected = np.array([
[0, 1, 2],
[MV, MV, 0],
[0, 1, 2]
])
np.testing.assert_equal(anno.annotations, expected)
def estimate_quality_for_q(annotations, qnum, pmids=None):
m, workers = get_M_q(annotations, qnum, pmids=pmids)
q_model = ModelB.create_initial_state(2, len(workers))
anno = AnnotationsContainer.from_array(m, missing_values=[2])
q_model.map(anno.annotations)
'''
pi[k] is the probability of label k
theta[j,k,k'] is the probability that
annotator j reports label k' for an
item whose real label is k, i.e.
P( annotator j chooses k' | real label = k)
'''
# this is a simple mean of sensitivity and specificity
# @TODO revisit?
proxy_skill = (q_model.theta[:, 0, 0] + q_model.theta[:, 1, 1]) / 2.0
return dict(zip(workers, proxy_skill))
def test_from_file_numerical_labels(self):
s = """
1 -1 2
2 1 3
-1 -1 3
"""
buffer = StringIO(s)
anno = AnnotationsContainer._from_file_object(buffer)
expected = np.array([[0, MV, 1], [1, 0, 2], [MV, MV, 2]])
np.testing.assert_equal(anno.annotations, expected)
self.assertEqual(anno.nclasses, 3)
self.assertEqual(anno.labels, ['1', '2', '3'])
self.assertEqual(anno.missing_values, ['-1'])
def test_from_file_string_labels(self):
s = """
D B A A B
C A * C A
B B D D *
"""
buffer = StringIO(s)
anno = AnnotationsContainer._from_file_object(buffer)
self.assertEqual(anno.nclasses, 4)
self.assertEqual(anno.nannotators, 5)
# labels should be sorted
self.assertEqual(anno.labels, ['A', 'B', 'C', 'D'])
self.assertEqual(anno.missing_values, ['*'])
expected = np.array(
[[3, 1, 0, 0, 1], [2, 0, MV, 2, 0], [1, 1, 3, 3, MV]], dtype=int)
np.testing.assert_equal(anno.annotations, expected)
def test_from_file_numerical_labels(self):
s = """
1 -1 2
2 1 3
-1 -1 3
"""
buffer = StringIO(s)
anno = AnnotationsContainer._from_file_object(buffer)
expected = np.array([
[0, MV, 1],
[1, 0, 2],
[MV, MV, 2]
])
np.testing.assert_equal(anno.annotations, expected)
self.assertEqual(anno.nclasses, 3)
self.assertEqual(anno.labels, ['1', '2', '3'])
self.assertEqual(anno.missing_values, ['-1'])
def test_from_file_string_labels(self):
s = """
D B A A B
C A * C A
B B D D *
"""
buffer = StringIO(s)
anno = AnnotationsContainer._from_file_object(buffer)
self.assertEqual(anno.nclasses, 4)
self.assertEqual(anno.nannotators, 5)
# labels should be sorted
self.assertEqual(anno.labels, ['A', 'B', 'C', 'D'])
self.assertEqual(anno.missing_values, ['*'])
expected = np.array([
[3, 1, 0, 0, 1],
[2, 0, MV, 2, 0],
[1, 1, 3, 3, MV]
], dtype=int)
np.testing.assert_equal(anno.annotations, expected)
def _create_new_entry(model, annotations, id):
value = model.log_likelihood(annotations)
ac = AnnotationsContainer.from_array(annotations, name=id)
db.store_result(id, ac, model, value)
def _create_new_entry(model, annotations, id):
value = model.log_likelihood(annotations)
ac = AnnotationsContainer.from_array(annotations, name=id)
db.store_result(id, ac, model, value)
def _annotations_view_default(self):
anno = AnnotationsContainer.from_array([[0]], name='<undefined>')
return AnnotationsView(annotations_container=anno,
nclasses=self.model.nclasses,
application=self.application,
model=HasTraits())
def _update_annotations_file(self):
logger.info('Load file {}'.format(self.annotations_file))
anno = AnnotationsContainer.from_file(self.annotations_file)
self.set_annotations(anno)
def _update_annotations_file(self):
logger.info('Load file {}'.format(self.annotations_file))
anno = AnnotationsContainer.from_file(self.annotations_file)
self.set_annotations(anno)
def _annotations_view_default(self):
anno = AnnotationsContainer.from_array([[0]], name='<undefined>')
return AnnotationsView(annotations_container = anno,
nclasses = self.model.nclasses,
application = self.application,
model=HasTraits())
