from utils.models import cnn_midfusion_model_v2, lstm_model
import sys
import math
import numpy as np
import os
if __name__ == "__main__":
dataset = sys.argv[1]
method = sys.argv[2]
proto_num = int(sys.argv[3])
os.environ["CUDA_VISIBLE_DEVICES"] = sys.argv[4]
max_seq_lenth = max_seq_len(dataset)
nb_class = nb_classes(dataset)
dim_num = nb_dims(dataset)
nb_cnn = int(round(math.log(max_seq_lenth, 2)) - 3)
#model = cnn_raw_model(nb_cnn, proto_num, max_seq_lenth, nb_class)
#model = cnn_dtwfeatures_model(nb_cnn, proto_num, max_seq_lenth, nb_class)
#model = cnn_earlyfusion_model(nb_cnn, proto_num, max_seq_lenth, nb_class)
model = cnn_midfusion_model_v2(nb_cnn, dim_num, proto_num, max_seq_lenth,
nb_class)
#model = cnn_latefusion_model(nb_cnn, proto_num, max_seq_lenth, nb_class)
print("Number of Pooling Layers: %s" % str(nb_cnn))
train_model(model,
dataset,
method,
proto_num,
def load_dataset_at(index,
method,
proto_num,
normalize_timeseries=False,
verbose=True) -> (np.array, np.array):
dim = nb_dims(index)
train_data1 = "data/all-raw-train-data-%s-%s-%s.txt" % (index, method,
str(proto_num))
test_data1 = "data/all-raw-test-data-%s-%s-%s.txt" % (index, method,
str(proto_num))
train_data2 = "data/all-dtw_features-train-data-%s-%s-%s.txt" % (
index, method, str(proto_num))
test_data2 = "data/all-dtw_features-test-data-%s-%s-%s.txt" % (
index, method, str(proto_num))
train_labels = "data/all-train-label-%s-%s-%s.txt" % (index, method,
str(proto_num))
test_labels = "data/all-test-label-%s-%s-%s.txt" % (index, method,
str(proto_num))
is_timeseries = True # assume all input data is univariate time series
if os.path.exists(train_data1):
df = pd.read_csv(train_data1,
delimiter=' ',
header=None,
encoding='latin-1')
else:
raise FileNotFoundError('File %s not found!' % (train_data1))
X_train1 = df.values
X_train1 = np.reshape(
X_train1,
(np.shape(X_train1)[0], dim, int(np.shape(X_train1)[1] / (dim))))
if normalize_timeseries:
X_train1_min = np.min(X_train1)
X_train1_max = np.max(X_train1)
X_train1 = 2. * (X_train1 - X_train1_min) / (X_train1_max -
X_train1_min) - 1.
if os.path.exists(train_data2):
df = pd.read_csv(train_data2,
delimiter=' ',
header=None,
encoding='latin-1')
else:
raise FileNotFoundError('File %s not found!' % (train_data2))
X_train2 = df.values
X_train2 = np.reshape(X_train2, (np.shape(X_train2)[0], proto_num,
int(np.shape(X_train2)[1] / (proto_num))))
if normalize_timeseries:
X_train2_min = np.min(X_train2)
X_train2_max = np.max(X_train2)
X_train2 = 2. * (X_train2 - X_train2_min) / (X_train2_max -
X_train2_min) - 1.
if os.path.exists(train_labels):
df = pd.read_csv(train_labels,
delimiter=' ',
header=None,
encoding='latin-1')
else:
raise FileNotFoundError('File %s not found!' % (train_labels))
y_train = df[[1]].values
no_classes = nb_classes(index) #len(np.unique(y_train))
if os.path.exists(test_data1):
df = pd.read_csv(test_data1,
delimiter=' ',
header=None,
encoding='latin-1')
else:
raise FileNotFoundError('File %s not found!' % (test_data1))
X_test1 = df.values
X_test1 = np.reshape(
X_test1,
(np.shape(X_test1)[0], dim, int(np.shape(X_test1)[1] / (dim))))
if normalize_timeseries:
X_test1 = 2. * (X_test1 - X_train1_min) / (X_train1_max -
X_train1_min) - 1.
if os.path.exists(test_data2):
df = pd.read_csv(test_data2,
delimiter=' ',
header=None,
encoding='latin-1')
else:
raise FileNotFoundError('File %s not found!' % (test_data2))
X_test2 = df.values
X_test2 = np.reshape(X_test2, (np.shape(X_test2)[0], proto_num,
int(np.shape(X_test2)[1] / (proto_num))))
if normalize_timeseries:
X_test2 = 2. * (X_test2 - X_train2_min) / (X_train2_max -
X_train2_min) - 1.
if os.path.exists(test_labels):
df = pd.read_csv(test_labels,
delimiter=' ',
header=None,
encoding='latin-1')
else:
raise FileNotFoundError('File %s not found!' % (test_labels))
y_test = df[[1]].values
if verbose:
print("Finished loading test dataset..")
print()
print("Number of train samples : ", X_train1.shape[0],
"Number of test samples : ", X_test1.shape[0])
print("Number of classes : ", no_classes)
print("Sequence length : ", X_train1.shape[-1])
return X_train1, X_train2, y_train, X_test1, X_test2, y_test, is_timeseries