def gaussin_embedding_ap(filename_feature, filename_list_key, filename_scaler):
"""I wanted to use the gaussian as the embedding,
but Bhattacharyya distance may not in [0, 1], so I normalized them by dividing the maximum value """
logger = logging.getLogger(__name__)
list_feature_flatten_test, label_integer_test, le, scaler = \
load_data_embedding(filename_feature=filename_feature,
filename_list_key=filename_list_key,
filename_scaler=filename_scaler)
embeddings = []
for ii_emb in range(len(list_feature_flatten_test)):
logger.info('calculating..., %s, total, %s', ii_emb,
len(list_feature_flatten_test))
feature = scaler.transform(list_feature_flatten_test[ii_emb])
# print(feature.shape)
gmm = GaussianMixture(n_components=1, covariance_type='diag')
gmm.fit(feature)
mu = gmm.means_[0, :]
cov = gmm.covariances_[0, :]
embeddings.append([mu, cov])
dist_mat = distance_matrix_embedding_gaussian(embeddings)
dist_mat = 1.0 - dist_mat / np.max(dist_mat)
gt_mat = ground_truth_matrix(label_integer_test)
ap = eval_embeddings(dist_mat=dist_mat, gt_mat=gt_mat)
return ap
def embedding_siamese_tsne(filename_feature, filename_list_key, filename_scaler):
"""calculate average precision of siamese triplet embedding"""
list_feature_flatten_test, label_integer_test, le, scaler = \
load_data_embedding(filename_feature=filename_feature,
filename_list_key=filename_list_key,
filename_scaler=filename_scaler)
path_model = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/models/phoneme_embedding_siamese_triplet'
path_eval = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/eval/phoneme_embedding_siamese_triplet'
model_name = 'phone_embedding_RNN_triplet_margin08'
np.save(file=os.path.join(path_eval, model_name + '_labels'), arr=label_integer_test)
pickle.dump(le, open(os.path.join(path_eval, model_name + '_le.pkl'), 'wb'), protocol=2)
embedding_dim = 29
for ii in range(1):
filename_model = os.path.join(path_model, model_name + '_' + str(ii) + '.h5')
model = load_model(filepath=filename_model, compile=False)
model_embedding = model.get_layer('embedding')
embeddings = np.zeros((len(list_feature_flatten_test), embedding_dim))
for ii_emb in range(len(list_feature_flatten_test)):
print('calculate', ii, 'run time', ii_emb, 'embedding', len(list_feature_flatten_test), 'total')
x_batch = np.expand_dims(scaler.transform(list_feature_flatten_test[ii_emb]), axis=0)
embeddings[ii_emb, :] = model_embedding.predict_on_batch(x_batch)
np.save(file=os.path.join(path_eval, model_name + '_embedding_' + str(ii)), arr=embeddings)
def embedding_siamese_ap(filename_feature, filename_list_key, filename_scaler,
model_name):
"""calculate average precision of siamese triplet embedding"""
list_feature_flatten_test, label_integer_test, le, scaler = \
load_data_embedding(filename_feature=filename_feature,
filename_list_key=filename_list_key,
filename_scaler=filename_scaler)
path_model = '/home/gong/Documents/pycharmProjects/phoneticSimilarity/models/phoneme_embedding_siamese_triplet'
path_eval = '/home/gong/Documents/pycharmProjects/phoneticSimilarity/eval/phoneme_embedding_siamese_triplet'
list_ap = []
embedding_dim = 29
for ii in range(5):
filename_model = os.path.join(path_model,
model_name + '_' + str(ii) + '.h5')
model = load_model(filepath=filename_model, compile=False)
model_embedding = model.get_layer('embedding')
embeddings = np.zeros((len(list_feature_flatten_test), embedding_dim))
for ii_emb in range(len(list_feature_flatten_test)):
print('calculate', ii, 'run time', ii_emb, 'embedding',
len(list_feature_flatten_test), 'total')
x_batch = np.expand_dims(scaler.transform(
list_feature_flatten_test[ii_emb]),
axis=0)
embeddings[ii_emb, :] = model_embedding.predict_on_batch(x_batch)
# dist_mat = distance_matrix_embedding_classifier(embeddings)
dist_mat = (2.0 - squareform(pdist(embeddings, 'cosine'))) / 2.0
gt_mat = ground_truth_matrix(label_integer_test)
np.save(file=os.path.join(path_eval, 'dist_mat_' + str(ii)),
arr=dist_mat)
ap = eval_embeddings(dist_mat=dist_mat, gt_mat=gt_mat)
list_ap.append(ap)
filename_eval = os.path.join(path_eval, model_name + '.csv')
with open(filename_eval, 'w') as csvfile:
csvwriter = csv.writer(
csvfile,
delimiter=',',
)
csvwriter.writerow([np.mean(list_ap), np.std(list_ap)])
def embedding_classifier_tsne(filename_feature, filename_list_key, filename_scaler):
"""calculate average precision of classifier embedding"""
list_feature_flatten_test, label_integer_test, le, scaler = \
load_data_embedding(filename_feature=filename_feature,
filename_list_key=filename_list_key,
filename_scaler=filename_scaler)
path_model = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/models/phone_embedding_classifier'
path_eval = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/eval/phone_embedding_classifier'
# configs = [[1, 0], [1, 1], [2, 0], [2, 1], [2, 2], [3, 0], [3, 1], [3, 2], [3, 3]]
configs = [[1, 1]]
for config in configs:
model_name = config_select(config)
pickle.dump(le, open(os.path.join(path_eval, model_name + '_le.pkl'), 'wb'), protocol=2)
embedding_dim = 29
for ii in range(1):
filename_model = os.path.join(path_model, model_name + '_' + str(ii) + '.h5')
model = load_model(filepath=filename_model)
weights = model.get_weights()
input_shape = [1, None, 80]
model_1_batch = model_select(config=config, input_shape=input_shape)
model_1_batch.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
model_1_batch.set_weights(weights=weights)
embeddings = np.zeros((len(list_feature_flatten_test), embedding_dim))
for ii_emb in range(len(list_feature_flatten_test)):
print('calculate', ii, 'run time', ii_emb, 'embedding', len(list_feature_flatten_test), 'total')
x_batch = np.expand_dims(scaler.transform(list_feature_flatten_test[ii_emb]), axis=0)
embeddings[ii_emb, :] = model_1_batch.predict_on_batch(x_batch)
np.save(file=os.path.join(path_eval, model_name + '_embedding_' + str(ii)), arr=embeddings)
np.save(file=os.path.join(path_eval, model_name + '_labels_'), arr=label_integer_test)
# filename_label_encoder = '/homedtic/rgong/phoneEmbeddingModelsTraining/dataset/le_phn_embedding.pkl'
# filename_data_splits = '/homedtic/rgong/phoneEmbeddingModelsTraining/dataset/data_splits.pkl'
#
# path_model = '/homedtic/rgong/phoneEmbeddingModelsTraining/out/'
filename_feature = '/Users/gong/Documents/MTG document/dataset/phoneEmbedding/feature_phn_embedding_train.pkl'
filename_list_key = '/Users/gong/Documents/MTG document/dataset/phoneEmbedding/list_key.pkl'
filename_scaler = '/Users/gong/Documents/MTG document/dataset/phoneEmbedding/scaler_phn_embedding.pkl'
filename_label_encoder = '/Users/gong/Documents/MTG document/dataset/phoneEmbedding/le_phn_embedding.pkl'
filename_data_splits = '/Users/gong/Documents/MTG document/dataset/phoneEmbedding/data_splits.pkl'
path_model = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/models/phone_embedding_classifier/'
# path_model = '../temp'
list_feature_flatten, labels_integer, le, scaler = load_data_embedding(
filename_feature=filename_feature,
filename_list_key=filename_list_key,
filename_scaler=filename_scaler)
# # split folds
# folds5_split_indices = cv5foldsIndices(list_feature_flatten=list_feature_flatten, label_integer=labels_integer)
# index_feature = range(len(list_feature_flatten))
# train_index, val_index, _, _ = train_test_split(index_feature, labels_integer, test_size=0.1, stratify=labels_integer)
#
# pickle.dump(le ,open(filename_label_encoder, 'wb'), protocol=2)
# pickle.dump([train_index, val_index], open(filename_data_splits, 'wb'), protocol=2)
train_index, val_index = pickle.load(open(filename_data_splits, 'rb'))
#
# for train_index, val_index in folds5_split_indices:
def embedding_classifier_ap(filename_feature, filename_list_key,
filename_scaler):
"""calculate average precision of classifier embedding"""
list_feature_flatten_test, label_integer_test, le, scaler = \
load_data_embedding(filename_feature=filename_feature,
filename_list_key=filename_list_key,
filename_scaler=filename_scaler)
path_model = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/models/phone_embedding_classifier'
path_eval = '/Users/gong/Documents/pycharmProjects/phoneticSimilarity/eval/phone_embedding_classifier'
# configs = [[1, 0], [1, 1], [2, 0], [2, 1], [2, 2], [3, 0], [3, 1], [3, 2], [3, 3]]
configs = [[1, 1]]
for config in configs:
model_name = config_select(config)
list_ap = []
embedding_dim = 29
for ii in range(5):
filename_model = os.path.join(path_model,
model_name + '_' + str(ii) + '.h5')
model = load_model(filepath=filename_model)
weights = model.get_weights()
input_shape = [1, None, 80]
model_1_batch = model_select(config=config,
input_shape=input_shape)
model_1_batch.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
model_1_batch.set_weights(weights=weights)
embeddings = np.zeros(
(len(list_feature_flatten_test), embedding_dim))
for ii_emb in range(len(list_feature_flatten_test)):
print('calculate', ii, 'run time', ii_emb, 'embedding',
len(list_feature_flatten_test), 'total')
x_batch = np.expand_dims(scaler.transform(
list_feature_flatten_test[ii_emb]),
axis=0)
embeddings[ii_emb, :] = model_1_batch.predict_on_batch(x_batch)
# dist_mat = distance_matrix_embedding_classifier(embeddings)
dist_mat = (2.0 - squareform(pdist(embeddings, 'cosine'))) / 2.0
gt_mat = ground_truth_matrix(label_integer_test)
np.save(file=os.path.join(path_eval, 'dist_mat_' + str(ii)),
arr=dist_mat)
ap = eval_embeddings(dist_mat=dist_mat, gt_mat=gt_mat)
list_ap.append(ap)
filename_eval = os.path.join(path_eval, model_name + '.csv')
with open(filename_eval, 'w') as csvfile:
csvwriter = csv.writer(
csvfile,
delimiter=',',
)
csvwriter.writerow([np.mean(list_ap), np.std(list_ap)])