def test_th(self, activate_k, k_set):
self.logger.info("Model testing thresholding starts")
self.logger.info("Activation k(={}) in embeddings(={})".format(
activate_k, self.args.m))
test_k_activate = activate_k_2D(self.test_embed,
k=activate_k,
session=self.sess) # [ntest, args.m]
train_k_activate = activate_k_2D(self.train_embed,
k=activate_k,
session=self.sess) # [ntrain, args.m]
self.regen_session()
if not hasattr(self, 'te_tr_distance') and not hasattr(
self, 'te_te_distance'):
self.regen_session()
self.te_tr_distance = self.sess.run(
pairwise_distance_euclid_v2(
tf.convert_to_tensor(self.test_embed, dtype=tf.float32),
tf.convert_to_tensor(self.train_embed, dtype=tf.float32)))
self.te_te_distance = self.sess.run(
pairwise_distance_euclid(
tf.convert_to_tensor(self.test_embed, dtype=tf.float32)))
performance = evaluate_hash_te_tr(train_hash_key=train_k_activate, test_hash_key=test_k_activate,\
te_tr_distance=self.te_tr_distance, te_te_distance=self.te_te_distance,\
te_tr_query_key=test_k_activate, te_tr_query_value=self.test_embed,\
te_te_query_key=test_k_activate, te_te_query_value=self.test_embed,\
train_label=self.train_label, test_label=self.test_label,\
ncls_train=self.ncls_train, ncls_test=self.ncls_test,\
activate_k=activate_k, k_set=k_set, logger=self.logger)
self.logger.info("Model testing thresholding ends")
return performance
def test1():
'''
test for activate_k_1D,
test for activate_k_2D
Results -
arr : [3 6 5 4]
activated arr : [0 1 1 0]
arr : [[3 6 5 4]
[2 1 3 4]]
activated arr : [[0 1 1 0]
[0 0 1 1]]
activated arr : [[0 1 1 0]
[0 0 1 1]]
arr : [[3 6 5 4]
[2 1 3 4]]
'''
arr = np.array([3, 6, 5, 4])
print("arr :", arr)
print("activated arr :", activate_k_1D(arr, 2))
arr = np.array([[3, 6, 5, 4], [2, 1, 3, 4]])
print("arr :", arr)
print("activated arr :", activate_k_2D(arr, 2))
with tf.Session() as sess:
print("activated arr :", activate_k_2D(arr, 2, sess))
print("arr : ", arr)
def test2():
# test for activate_k_2D
arr = np.array([[3, 6, 5, 4], [2, 1, 3, 4]])
print("arr :", arr)
print("activated arr :", activate_k_2D(arr, 2))
# test for activate_k_2D_fast
with tf.Session() as sess:
print("activated arr :", activate_k_2D(arr, 2, sess))
print("arr : ", arr)
def test_vq(self, activate_k, k_set):
self.logger.info("Model testing vq starts")
self.logger.info("Activation k(={}) in buckets(={})".format(
activate_k, self.args.m))
if not hasattr(self, 'train_kmc') and not hasattr(self, 'test_kmc'):
self.test_kmc = KMeansClustering(self.test_embed, self.args.m)
self.train_kmc = KMeansClustering(self.train_embed, self.args.m)
self.regen_session()
if not hasattr(self, 'te_tr_distance') and not hasattr(
self, 'te_te_distance'):
self.te_tr_distance = self.sess.run(
pairwise_distance_euclid_v2(
tf.convert_to_tensor(self.test_embed, dtype=tf.float32),
tf.convert_to_tensor(self.train_embed, dtype=tf.float32)))
self.te_te_distance = self.sess.run(
pairwise_distance_euclid(
tf.convert_to_tensor(self.test_embed, dtype=tf.float32)))
self.regen_session()
te_te_query_value = self.test_kmc.k_hash(
self.test_embed, self.sess) # [ntest, args.m] center test
te_tr_query_value = self.train_kmc.k_hash(
self.test_embed, self.sess) # [ntest, args.m] center train
self.regen_session()
te_te_query_key = activate_k_2D(te_te_query_value,
k=activate_k,
session=self.sess) # [ntest, args.m]
test_hash_key = te_te_query_key
te_tr_query_key = activate_k_2D(te_tr_query_value,
k=activate_k,
session=self.sess) # [ntest, args.m]
train_hash_key = activate_k_2D(self.train_kmc.k_hash(
self.train_embed, self.sess),
k=activate_k,
session=self.sess) # [ntrain, args.m]
self.regen_session()
performance = evaluate_hash_te_tr(train_hash_key=train_hash_key, test_hash_key=test_hash_key,\
te_tr_distance=self.te_tr_distance, te_te_distance=self.te_te_distance,\
te_tr_query_key=te_tr_query_key, te_tr_query_value=te_tr_query_value,\
te_te_query_key=te_te_query_key, te_te_query_value=te_te_query_value,\
train_label=self.train_label, test_label=self.test_label,\
ncls_train=self.ncls_train, ncls_test=self.ncls_test,\
activate_k=activate_k, k_set=k_set, logger=self.logger)
self.logger.info("Model testing vq ends")
return performance
def test_hash_metric(self, activate_k, k_set):
self.logger.info("Model testing k hash starts")
self.logger.info("Activation k(={}) in buckets(={})".format(
activate_k, self.args.d))
self.regen_session()
test_k_activate = activate_k_2D(self.test_k_hash,
k=activate_k,
session=self.sess) # [ntest, args.d]
if not hasattr(self, 'te_te_distance'):
self.regen_session()
self.te_te_distance = pairwise_distance_euclid_efficient(
input1=self.test_embed,
input2=self.test_embed,
session=self.sess,
batch_size=128)
self.logger.info(
"Calculating pairwise distance from test embeddings")
performance = evaluate_hash_te(test_hash_key=test_k_activate, te_te_distance=self.te_te_distance,\
te_te_query_key=test_k_activate, te_te_query_value=self.test_k_hash,\
test_label=self.test_label, ncls_test=self.ncls_test,\
activate_k=activate_k, k_set=k_set, logger=self.logger)
self.logger.info("Model testing k hash ends")
return performance
def test_vq(self, activate_k, k_set):
self.logger.info("Model testing vq starts")
self.logger.info("Activation k(={}) in buckets(={})".format(
activate_k, self.args.m))
if not hasattr(self, 'test_kmc'):
self.regen_session()
self.test_kmc = KMeansClustering(self.test_embed, self.args.m)
if not hasattr(self, 'te_te_distance'):
self.regen_session()
self.te_te_distance = pairwise_distance_euclid_efficient(
input1=self.test_embed,
input2=self.test_embed,
session=self.sess,
batch_size=128)
self.logger.info(
"Calculating pairwise distance from test embeddings")
te_te_query_value = self.test_kmc.k_hash(
self.test_embed, self.sess) # [ntest, args.d] center test
te_te_query_key = activate_k_2D(te_te_query_value,
k=activate_k,
session=self.sess) # [ntest, args.d]
test_hash_key = te_te_query_key
self.regen_session()
performance = evaluate_hash_te(test_hash_key=test_hash_key, te_te_distance=self.te_te_distance,\
te_te_query_key=te_te_query_key, te_te_query_value=te_te_query_value,\
test_label=self.test_label, ncls_test=self.ncls_test,\
activate_k=activate_k, k_set=k_set, logger=self.logger)
self.logger.info("Model testing vq ends")
return performance