def test_that_retrain_calls_eval_method_correctly(mocker):
mock_classifier = mocker.Mock()
mock_classifier.fit == mocker.Mock()
mock_classifier.predict_proba == mocker.Mock()
mock_eval_method = mocker.Mock(
return_value={"test_score": np.array([0.8])}
)
test_array = np.array([[1, 2, 3], [1, 2, 3], [1, 2, 3]])
widget = SemiSupervisor(
features=test_array,
classifier=mock_classifier,
eval_method=mock_eval_method,
)
widget._annotation_loop.send({"source": "", "value": "dummy label 1"})
widget._annotation_loop.send({"source": "", "value": "dummy label 2"})
widget.retrain()
assert mock_eval_method.call_count == 1
call_arguments = mock_eval_method.call_args[0]
assert call_arguments[0] is mock_classifier
assert (call_arguments[1] == test_array[:2, :]).all()
assert pytest.helpers.same_elements(
call_arguments[2], ["dummy label 1", "dummy label 2"]
)
assert widget.model_performance.value == "Score: 0.80"
def test_that_retrain_calls_reorder_correctly(mocker):
test_probabilities = np.array([[0.2, 0.3], [0.1, 0.4]])
mock_eval_method = mocker.Mock(
return_value={"test_score": np.array([0.8])}
)
mock_reordering = mocker.Mock(return_value=[0, 1])
test_array = np.array([[1, 2, 3], [1, 2, 3], [1, 2, 3], [1, 2, 3]])
widget = SemiSupervisor(
features=test_array,
classifier=LogisticRegression(),
eval_method=mock_eval_method,
reorder=mock_reordering,
shuffle_prop=0.2,
)
mocker.patch.object(
widget.classifier, "fit", return_value=LogisticRegression()
)
mocker.patch.object(
widget.classifier, "predict_proba", return_value=test_probabilities
)
widget._annotation_loop.send({"source": "", "value": "dummy label 1"})
widget._annotation_loop.send({"source": "", "value": "dummy label 2"})
widget.retrain()
assert mock_reordering.call_count == 1
call_args, call_kwargs = mock_reordering.call_args_list[0]
assert (call_args[0] == test_probabilities).all()
assert call_kwargs["shuffle_prop"] == 0.2
def test_that_retrain_with_no_labels_sets_warnings(mocker):
mock_classifier = mocker.Mock()
mock_classifier.fit == mocker.Mock()
mock_classifier.predict_proba == mocker.Mock()
mock_eval_method = mocker.Mock(
return_value={"test_score": np.array([0.8])}
)
test_array = np.array([[1, 2, 3], [1, 2, 3], [1, 2, 3]])
widget = SemiSupervisor(
features=test_array,
classifier=mock_classifier,
eval_method=mock_eval_method,
)
widget.retrain()
assert (
widget.model_performance.value
== "Score: Not enough labels to retrain."
)
widget._annotation_loop.send({"source": "", "value": "dummy label 1"})
widget.retrain()
assert (
widget.model_performance.value
== "Score: Not enough labels to retrain."
)
widget._annotation_loop.send({"source": "", "value": "dummy label 2"})
widget.retrain()
assert mock_eval_method.call_count == 1
assert widget.model_performance.value == "Score: 0.80"
def test_that_calling_retrain_without_classifier_breaks():
with pytest.raises(ValueError):
test_array = np.array([[1, 2, 3], [1, 2, 3], [1, 2, 3]])
widget = SemiSupervisor(features=test_array)
widget.retrain()
