def test_estimate_semantic_data_cub_ae(self):
"""
Tests if result from semantic data estimation is in the correct shape for cub
"""
data = loadmat('../../../../Datasets/SEM/cub_demo_data.mat')
input_length = output_length = data['X_tr'].shape[1] + data[
'S_tr'].shape[1]
ae = Encoder(input_length, data['S_tr'].shape[1], output_length,
ModelType.SIMPLE_AE_1L, 1)
labels = data['te_cl_id']
labels_dict = {
labels[i][0]: attributes
for i, attributes in enumerate(data['S_te_pro'])
}
s_te = np.array(
[labels_dict[label[0]] for label in data['test_labels_cub']])
sem_tr, sem_te = ae.estimate_semantic_data(data['X_tr'], data['X_te'],
data['S_tr'], s_te)
self.assertEqual((data['X_tr'].shape[0], data['S_tr'].shape[1]),
sem_tr.shape)
self.assertEqual((data['X_te'].shape[0], data['S_tr'].shape[1]),
sem_te.shape)
def estimate_sem_data_sec(data):
tr_vis_data = normalize(data['X_tr'], norm='l2', axis=1, copy=True)
tr_sem_data = normalize(data['S_tr'], norm='l2', axis=1, copy=True)
te_vis_data = normalize(data['X_te'], norm='l2', axis=1, copy=True)
input_length = output_length = tr_vis_data.shape[1] + tr_sem_data.shape[1]
ae = Encoder(input_length, tr_sem_data.shape[1], output_length, ModelType.ZSL_AE, 50, '.', run_svm=False)
tr_sem, te_sem = ae.estimate_semantic_data_zsl(tr_vis_data, te_vis_data, tr_sem_data, False)
return tr_sem, te_sem
def estimate_sem_data(tr_vis_data, te_vis_data, tr_sem_data, res_path):
tr_sem_data = normalize(tr_sem_data, norm='l2', axis=1, copy=True)
tr_vis_data = normalize(tr_vis_data, norm='l2', axis=1, copy=True)
te_vis_data = normalize(te_vis_data, norm='l2', axis=1, copy=True)
input_length = output_length = tr_vis_data.shape[1] + tr_sem_data.shape[1]
ae = Encoder(input_length, tr_sem_data.shape[1], output_length,
ModelType.ZSL_AE, 50, res_path)
tr_sem, te_sem = ae.estimate_semantic_data_zsl(tr_vis_data, te_vis_data,
tr_sem_data, False)
return tr_sem, te_sem
def setUpClass(cls):
"""
Builds and train AE model
"""
data = loadmat('../../../../Datasets/SEM/cub_demo_data.mat')
input_length = output_length = data['X_tr'].shape[1] + data[
'S_tr'].shape[1]
cls.ae = Encoder(input_length, data['S_tr'].shape[1], output_length,
ModelType.SIMPLE_AE, 5)
labels = data['te_cl_id']
labels_dict = {
labels[i][0]: attributes
for i, attributes in enumerate(data['S_te_pro'])
}
s_te = np.array(
[labels_dict[label[0]] for label in data['test_labels_cub']])
cls.ae.estimate_semantic_data(data['X_tr'], data['X_te'], data['S_tr'],
s_te)
def estimate_vse_data(self, tr_vis, te_vis, tr_sem, te_sem, y_train,
y_test, res_path):
tr_sem = normalize(tr_sem, norm='l2', axis=1, copy=True)
tr_vis = normalize(tr_vis, norm='l2', axis=1, copy=True)
te_vis = normalize(te_vis, norm='l2', axis=1, copy=True)
te_sem = normalize(te_sem, norm='l2', axis=1, copy=True)
te_sem = SemanticDegradation.kill_semantic_attributes(
te_sem, self.degradation_rate)
te_sem = normalize(te_sem, norm='l2', axis=1, copy=True)
input_length = output_length = tr_vis.shape[1] + tr_sem.shape[1]
ae = Encoder(input_length, tr_sem.shape[1], output_length,
ModelType.STRAIGHT_AE, self.epochs, res_path,
self.run_svm)
tr_sem, te_sem = ae.estimate_semantic_data(tr_vis, te_vis, tr_sem,
te_sem, y_train, y_test,
self.save_results)
return tr_sem, te_sem