def test_classifer(dataset,
classifier_cls,
dist_func=distance.euclidean,
dist_func_params=None):
print("=" * 80)
print("\n")
print("testing ", classifier_cls.__name__)
print("=" * 80)
# load data
file_train = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
# z-normalization
X_train = data_parser.z_normalize(X_train)
X_test = data_parser.z_normalize(X_test)
print("=" * 60)
print("basic information:")
print("file_train", file_train)
print("file_test", file_test)
print("X_train", X_train.shape)
print("y_train", y_train.shape)
print("X_test", X_test.shape)
print("y_test", y_test.shape)
print("=" * 60)
print("=" * 60)
print("knn classification ")
# set 1-nn classifier
n_jobs = 10
n_neighbors = 1
classifier = classifier_cls(n_neighbors=n_neighbors,
distfunc=dist_func,
distfunc_params=dist_func_params,
n_jobs=n_jobs)
print("=" * 60)
print("classifier parameters")
print("distance function: ", classifier.distfunc.__name__)
print("distance parameters: ", classifier.distfunc_params)
print("n_jobs: ", classifier.n_jobs)
print("n_neighbors: ", classifier.n_neighbors)
print("=" * 60)
time_start = time()
classifier.fit(X_train, y_train)
y_pred = classifier.predict(X_test)
report = classification_report(y_test, y_pred)
print(report)
time_end = time()
acc = validation.cal_accuracy(y_test, y_pred)
print("time: ", time_end - time_start)
print("accuracy: ", acc)
def experiment_cnn_shapelet(dataset, k=5, k_val=1):
file_train = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
# shapelet transformation
min_shapelet_length = 8
max_shapelet_length = 30
num_shapelet = int(X_train.shape[1] * 0.5)
length_increment = 10
position_increment = 20
callSTS = ShapeletTransformBasic(n_shapelets=num_shapelet,
min_shapelet_length=min_shapelet_length,
max_shapelet_length=max_shapelet_length,
length_increment=length_increment,
position_increment=position_increment)
print("shapelet learning.....")
callSTS.fit(X_train, y_train)
bestk_shapelets = callSTS.train()
print("bestk_shapelets: ", np.shape(bestk_shapelets))
print("shapelet transformation.....")
X_train_ = callSTS.transform(X_train)
X_test_ = callSTS.transform(X_test)
X_train_ = np.array(X_train_)
X_test_ = np.array(X_test_)
# cnn classifier
X_train_, y_train, X_val_, y_val = data_parser.k_fold_validation_balance(X_train_, y_train, k, k_val)
acc = cnn_classifier(X_train_, y_train, X_test_, y_test, X_val_, y_val)
# print result
print("=" * 80)
print("\n")
print("original data set: ")
print("train set", np.shape(X_train), np.shape(y_train))
print("test set", np.shape(X_test), np.shape(y_test))
print("shapelet transformed set: ")
print("learned shapelets: ", np.shape(bestk_shapelets))
print("train set: ", np.shape(X_train_), np.shape(y_train))
print("validation set: ", np.shape(X_val_), np.shape(y_val))
print("test set: ", np.shape(X_test_), np.shape(y_test))
print("\n")
print("shapelets learning parameter: ")
print("minShapeletLength", min_shapelet_length)
print("max_shapelet_length", max_shapelet_length)
print("length_increment", length_increment)
print("position_increment", position_increment)
print("\n")
print(acc)
print("=" * 80)
return acc
def test_find_length():
dataset = 'Beef'
file_train = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
print("estimating min length and max length of shapelet learning...")
min_length, max_length = estimate_min_max_length(X_train, y_train)
print(min_length)
print(max_length)
def test_k_fold_split():
print("=" * 80)
print("\n")
file_train = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
k = 5
k_val = 1
X_train, y_train, X_val, y_val = data_parser.k_fold_validation_balance(
X_train, y_train, k, k_val)
print(X_train.shape, y_train.shape)
print(X_val.shape, y_val.shape)
print("=" * 80)
print("\n")
def experiment_cnn(dataset, k=5, k_val=1):
file_train = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, dataset, dataset + '_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
X_train, y_train, X_val, y_val = data_parser.k_fold_validation_balance(X_train, y_train, k, k_val)
acc = cnn_classifier(X_train, y_train, X_test, y_test, X_val, y_val)
print("="*80)
print("\n")
print("data set: ")
print("train set", np.shape(X_train), np.shape(y_train))
print("validation set", np.shape(X_val), np.shape(y_val))
print("test set", np.shape(X_test), np.shape(y_test))
print(acc)
print("=" * 80)
return acc
def run(dataset,
shapelet_cls,
min_shapelet_length=0.1,
max_shapelet_length=0.7,
num_shapelet=0.5,
length_increment=20,
position_increment=20,
n_neighbors=1,
n_jobs=10,
is_normalize=False,
log_dir=None):
print("="*80)
print("\n")
print("testing ", shapelet_cls.__name__)
print("\n")
# load data
file_train = os.path.join(UCR_DATA_ROOT, dataset, dataset+'_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, dataset, dataset+'_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
# z normalization
if is_normalize:
X_train = data_parser.z_normalize(X_train)
X_test = data_parser.z_normalize(X_test)
print("="*80)
print("basic information: \n")
print("file_train: ", file_train)
print("file_test: ", file_test)
print("X_train", X_train.shape)
print("y_train", y_train.shape)
print("X_test", X_test.shape)
print("y_test", y_test.shape)
print("="*80)
# shapelet setup
if min_shapelet_length < 1:
min_shapelet_length = int(X_train.shape[1] * min_shapelet_length)
if max_shapelet_length < 1:
max_shapelet_length = int(X_train.shape[1] * max_shapelet_length)
if num_shapelet:
num_shapelet = int(X_train.shape[1] * num_shapelet)
callSTS = shapelet_cls(n_shapelets=num_shapelet,
min_shapelet_length=min_shapelet_length,
max_shapelet_length=max_shapelet_length,
length_increment=length_increment,
position_increment=position_increment)
callSTS.fit(X_train, y_train)
print("="*80)
print("set up shapelet learning and transformation: \n")
print("series_list", np.shape(callSTS.series_list))
print("class_list", np.shape(callSTS.class_list))
print("num shapelets", callSTS.n_shapelets)
print("min_shapelet_length", callSTS.min_shapelet_length)
print("max_shapelet_length", callSTS.max_shapelet_length)
print("length_increment", callSTS.length_increment)
print("position_increment", callSTS.position_increment)
print("distance function", str(callSTS.dist_func.__name__))
print("distance parameter", callSTS.dist_func_params)
if hasattr(callSTS, 'class_distribution'):
print("class distribution", callSTS.class_distribution)
print("="*80)
print("="*80)
print("shapelet learning ......")
_ = callSTS.train()
print("best k shapelet: ", np.shape(callSTS.best_shapelets_content_))
print("="*80)
print("="*80)
print("shapelet transformation ......")
X_train_transform = callSTS.transform(X_train)
X_test_transform = callSTS.transform(X_test)
X_train_transform = np.array(X_train_transform)
X_test_transform = np.array(X_test_transform)
print("X_train_transform: ", np.shape(X_train_transform))
print("X_test_transform: ", np.shape(X_test_transform))
# knn classification
print("="*80)
print("knn classification ......")
KNNC = KNeighborsClassifier(n_neighbors=n_neighbors, n_jobs=n_jobs)
KNNC.fit(X_train_transform, y_train)
y_pred = KNNC.predict(X_test_transform)
acc = validation.cal_accuracy(y_test, y_pred)
print("accuracy: ", acc)
# log shapelet
if log_dir is not None:
file_shapelet = os.path.join(log_dir,
"%s_minlen-%s_maxlen-%s" % (dataset, min_shapelet_length, max_shapelet_length))
print("saving shapelet to %s........." % file_shapelet)
callSTS.save_shapelet(file_shapelet)
return acc, callSTS.best_shapelets_content_
n_neighbors = 1
n_jobs = 10
k_fold = 10
distfunc = distance.euclidean
# set up knn classifier
knn_clf = KNeighborsClassifier(n_neighbors=n_neighbors,
n_jobs=n_jobs,
distfunc=distfunc)
result = [['dataset name', 'accuracy']]
for name in dataset_:
# load data set
file_train = os.path.join(UCR_DATA_ROOT, name, name + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, name, name + '_TEST')
X_train, y_train = data_parser.load_ucr(file_train)
X_test, y_test = data_parser.load_ucr(file_test)
# normalize
X_train = data_parser.z_normalize(X_train)
X_test = data_parser.z_normalize(X_test)
# k fold validation
acc_sum = 0
for i in range(k_fold):
knn_clf.fit(X_train, y_train)
y_pred = knn_clf.predict(X_test)
acc_sum += validation.cal_accuracy(y_test, y_pred)
X_train, y_train, X_test, y_test = data_parser.resample_data(X_train, y_train, X_test, y_test)
acc = acc_sum / k_fold
result.append((name, acc))
'dist_func': dist_func,
'dist_func_params': dist_func_params
}
print("=" * 80)
print(__doc__)
print("\n")
result = []
result.append(("data set name", "time", "time of parallel", "size",
"length of sub sequence", "length of distance list"))
for set_name in dataset:
file_train = os.path.join(UCR_DATA_ROOT, set_name, set_name + '_TRAIN')
file_test = os.path.join(UCR_DATA_ROOT, set_name, set_name + '_TEST')
X_train, _ = data_parser.load_ucr(file_train)
X_test, _ = data_parser.load_ucr(file_test)
X = np.vstack([X_train, X_test])
m, n = X.shape
sublen = int(0.5 * n)
subseq = X[0][:sublen]
t_start = time()
distance_list = []
for i in range(m):
dist = distance.dist_subsequence(subsequence=subseq,
wholeseries=X[i],
**params)
distance_list.append(dist)
t_end = time()
t_non_parallel = t_end - t_start