unlabeled_truth_set = set([tuple(x) for x in unlabeled_truth])
# Cotraining label prediction accuracy
avg_nn_pred_confidence.append(cotrain_model.avg_nn_pred_confidence)
avg_svm_pred_confidence.append(cotrain_model.avg_svm_pred_confidence)
cotrain_unlabeled_predictions = cotrain_model.get_unlabeled_predictions()
cotrain_unlabeled_predictions_set = set([tuple(x) for x in cotrain_unlabeled_predictions])
cotrain_correct_matches = np.array([x for x in cotrain_unlabeled_predictions_set & unlabeled_truth_set])
cotrain_label_acc = len(cotrain_correct_matches) / len(cotrain_unlabeled_predictions)
print('cotraining labeling accuracy: %s' % cotrain_label_acc)
print()
cotrain_unpredicted_data = cotrain_model.get_unpredicted_data()
kmeans = SemiSupervisedKMeans(num_clusters=2)
kmeans.initialize(training_data, training_labels)
kmeans.fit(unlabeled_data, 4500)
# Clustering prediction accuracy
cluster_unlabeled_predictions = kmeans.get_unlabeled_predictions()
cluster_unlabeled_predictions_set = set([tuple(x) for x in cluster_unlabeled_predictions])
cluster_correct_matches = np.array([x for x in cluster_unlabeled_predictions_set & unlabeled_truth_set])
cluster_label_acc = len(cluster_correct_matches) / len(cluster_unlabeled_predictions)
print('clustering labeling accuracy: %s' % cluster_label_acc)
print()
labeled_data = np.insert(training_data, 0, training_labels, axis=1)
remaining_unlabeled_predictions = kmeans.predict(cotrain_unpredicted_data, 4500)
complete_training_data = np.vstack([labeled_data, cotrain_unlabeled_predictions, remaining_unlabeled_predictions])
# Remaining data labeling accuracy
remaining_unlabeled_predictions_set = set([tuple(x) for x in remaining_unlabeled_predictions])
sample = 0
kfold = StratifiedKFold(n_splits=10, shuffle=True)
while sample < total_runs:
try:
training_data = df.values
training_labels = training_data[:, 0]
training_data = np.delete(training_data, 0, 1)
unlabeled_data = udf.values
unlabeled_labels = unlabeled_data[:, 0]
unlabeled_data = np.delete(unlabeled_data, 0, 1)
kmeans = SemiSupervisedKMeans(num_clusters=2)
kmeans.initialize(training_data, training_labels)
kmeans.fit(unlabeled_data, 4)
# Label prediction accuracy setup
unlabeled_truth = np.insert(unlabeled_data,
0,
unlabeled_labels,
axis=1)
unlabeled_truth_set = set([tuple(x) for x in unlabeled_truth])
# Clustering prediction accuracy
cluster_unlabeled_predictions = kmeans.get_unlabeled_predictions()
cluster_unlabeled_predictions_set = set(
[tuple(x) for x in cluster_unlabeled_predictions])
cluster_correct_matches = np.array([
x for x in cluster_unlabeled_predictions_set & unlabeled_truth_set
])