default=datasets.TYPES)
parser.add_argument(
'--ignore_missing',
help=
'Ignore functions from the dataset which no detections are present for',
action='store_true')
parser.add_argument(
'--num_samples',
help='Number of samples for permutation and bootstrap test',
type=int,
default=1000)
args = parser.parse_args()
# Load dataset
data = datasets.loadDatasets(args.datasets, args.extremetypes)
# Load detections
det1 = loadDetectionDump(args.dump1)
det2 = loadDetectionDump(args.dump2)
# Associate detections with each other and with ground-truth
det = alignDetections(det1,
det2,
data,
not args.ignore_missing,
quiet=False)
# Run significance tests and print results
results = testSignificance(det, args.num_samples, printProgress=True)
printSignificanceResults(results)
parser.add_argument('--subsets',
help='subsets of the datasets to be tested',
nargs='+',
default=[])
parser.add_argument('--td_lag',
help='Time-Lag for Time-Delay Embedding',
default=1,
type=int)
parser.add_argument(
'--dump',
help='Dump detections for each time-series to the specified CSV file',
default='')
args = parser.parse_args()
# Load data
data = datasets.loadDatasets(args.datasets)
ftypes = args.subsets if len(args.subsets) > 0 else data.keys()
# Find the best embedding dimension for every single time series
aucs = {}
aps = {}
all_ids = []
all_gt = []
all_regions = []
best_k = {}
for ftype in ftypes:
print('-- {} --'.format(ftype))
func_ids = []
ygts = []
regions = []
PROPOSALS = sys.argv[2] if len(sys.argv) > 2 else 'dense'
TD_DIM = int(sys.argv[3]) if len(sys.argv) > 3 else 6
TD_LAG = int(sys.argv[4]) if len(sys.argv) > 4 else 2
if PROPOSALS == 'help':
print(
'Test several deseasonalization methods in combination with different variants of the MDI algorithm on a given dataset.\n'
)
print(
'Usage: {} <dataset = yahoo | synthetic> <proposals = dense> <td-dim = 6> <td-lag = 2>'
.format(sys.argv[0]))
exit()
# Load test data
if DATASET == 'synthetic':
data = datasets.loadDatasets('synthetic_seasonal')['diurnal']
else:
data = datasets.loadDatasets('yahoo_real')['A1Benchmark']
# Check libmaxdiv
if libmaxdiv_wrapper.libmaxdiv is None:
raise RuntimeError('libmaxdiv could not be found and loaded.')
# Compile pipelines
pipelines = OrderedDict()
pipelines['none'] = []
pipelines['OLS'] = []
pipelines['Z-Score'] = []
params = libmaxdiv_wrapper.maxdiv_params_t()
libmaxdiv_wrapper.libmaxdiv.maxdiv_init_params(params)
params.min_size[0] = 20 if DATASET == 'synthetic' else 10
from maxdiv import preproc, eval
from maxdiv.baselines_noninterval import pointwiseRegionProposals
import datasets
# Constants
PROPMETHODS = ['hotellings_t', 'kde']
THS = np.concatenate((np.linspace(0, 2, 20, endpoint = False), np.linspace(2, 4, 9, endpoint = True)))
# Parse parameters
dataset = sys.argv[1] if len(sys.argv) > 1 else 'synthetic'
extint_max_len = max(10, int(sys.argv[2])) if len(sys.argv) > 2 else 100
td_dim = max(1, int(sys.argv[3])) if len(sys.argv) > 3 else 1
td_lag = max(1, int(sys.argv[4])) if len(sys.argv) > 4 else 1
# Load test data
data = datasets.loadDatasets(dataset, 'interval')
# Try different thresholds for interval proposing
results = OrderedDict()
for propmeth in PROPMETHODS:
results[propmeth] = OrderedDict()
for sd_th in THS:
ygts = []
regions = []
for ftype in data:
for func in data[ftype]:
ygts.append(func['gt'])
ts = preproc.normalize_time_series(func['ts'])
if td_dim > 1:
ts = preproc.td(ts, td_dim, td_lag)
action='store_true')
args = parser.parse_args()
# Prepare parameters for calling maxdiv
args_dict = vars(args)
parameters = {
parameter_name: args_dict[parameter_name]
for parameter_name in cli_tools.get_algorithm_parameters()
if (parameter_name in args_dict) and (parameter_name != 'td_dim')
}
if ('num_intervals' in parameters) and (parameters['num_intervals'] <= 0):
parameters['num_intervals'] = None
# Load synthetic test data
data = datasets.loadDatasets('synthetic')
# Determine set of extreme types to run tests for
extremetypes = set(args.extremetypes) & set(data.keys())
if len(extremetypes) == 0:
extremetypes = data.keys()
# Try all combinations of preprocessing methods, density estimators and divergence modes
# for all types of extremes and store the results in dictionaries
auc = {} # Area under ROC curve
auc_sd = {} # Standard deviation of AUC scores
aps = {} # Average Precision
times = {
} # Lists of runtimes for time series of different length and each method
labels = {} # Labels for the different combinations
all_gt = {}
PROPMETHODS = ['hotellings_t', 'kde']
METHOD = 'gaussian_cov'
MODE = 'CROSSENT'
MAD = [True, False]
FILTERED = [True, False]
THS = np.concatenate(
(np.linspace(0, 2, 20, endpoint=False), np.linspace(2, 4, 9,
endpoint=True)))
propmeth = sys.argv[1] if (len(sys.argv) > 1) and (
sys.argv[1] in PROPMETHODS) else PROPMETHODS[0]
dataset = sys.argv[2] if len(sys.argv) > 2 else 'synthetic'
# Load test data
data = datasets.loadDatasets(dataset)
# Try different parameter combinations for interval proposing
ap = OrderedDict() # Average Precision
mean_ap = OrderedDict() # Mean Average Precision
labels = OrderedDict() # Labels for the different combinations
for filtered in FILTERED:
for useMAD in MAD:
id = (filtered, useMAD)
ap[id] = {}
mean_ap[id] = {}
labels[id] = '{}, {}'.format('median' if useMAD else 'mean',
'gradients' if filtered else 'scores')
print('Testing {}'.format(labels[id]))
sys.stdout.flush()
