def test_four(self):
labels, truth = create_sim_dataset(1000, 6, 3, 0.8, 0.8, True, False,
False)
MLE = MLEOneParameterPerLabeler()
ll_list = []
results_list = []
for i in range(5):
results_list.append(MLE.fit(labels))
ll_list.append(
MLE.expert_models.log_likelihood(
labels, results_list[i]['class_names']))
index = np.argmax(np.asarray(ll_list))
accuracy = results_list[index]['accuracies']
self.assertTrue(np.all(np.abs(accuracy - truth) < 0.075))
def no_oracle(y_noisy, p_classifier, y_vetted, X=None):
y_classifier = np.argmax(p_classifier, axis=1)[:, np.newaxis]
if y_noisy.ndim == 1:
y_noisy = y_noisy.reshape(-1, 1)
y_vetted = y_vetted.reshape(-1, 1)
all_labels = np.hstack((y_classifier, y_noisy, y_vetted))
all_labels = np.asarray(all_labels, dtype=np.object)
all_labels[all_labels == -1] = np.nan
results = MLEOneParameterPerLabeler(n_classes=np.size(p_classifier, axis=1)).fit(all_labels)
return results['probs']
def test_return_fields(self):
mle_obj = MLEOneParameterPerLabeler()
result = mle_obj.fit(MED_MATRIX)
self.assertEqual(set(result.keys()), FIT_RETURN_SET)