Algorithms Overview and Metadata

We collected 158 time series anomaly detection algorithms from different research areas; Deep Learning, Statistics, Outlier Detection, Signal Analysis, Classic Machine Learning, Data Mining, Stochastic Learning. Some of these algorithms can detect anomalies on multidimensional time series.

Overview

We implemented 71 of the total collection. These implemented algorithms are used in our evaluation. The following table shows all categorized algorithms. Column Implemented shows which algorithms we used in the evaluation.

	Research Area	Method data type	Implemented
Method Name
AE	Deep Learning	multivariate	True
ARIMA	Statistics (Regression & Forecasting)	univariate	True
Bagel	Deep Learning	univariate	True
CBLOF	Ourlier Detection	multivariate	True
COF	Ourlier Detection	multivariate	True
COPOD	Ourlier Detection	multivariate	True
DAE	Deep Learning	multivariate	True
DBStream	Ourlier Detection	multivariate	True
DSPOT	Statistics (Regression & Forecasting)	univariate	True
DWT-MLEAD	Signal Analysis	univariate	True
DeepAnT	Deep Learning	multivariate	True
DeepNAP	Deep Learning	multivariate	True
Donut	Deep Learning	univariate	True
EIF	Classic Machine Learning	multivariate	True
EncDec-AD	Deep Learning	multivariate	True
Ensemble GI	Data Mining	univariate	True
FFT	Signal Analysis	univariate	True
Fast-MCD	Statistics (Regression & Forecasting)	multivariate	True
GrammarViz	Data Mining	univariate	True
HBOS	Classic Machine Learning	multivariate	True
HIF	Ourlier Detection	multivariate	True
HOT SAX	Data Mining	univariate	True
HealthESN	Deep Learning	univariate	True
Hybrid KNN	Deep Learning	multivariate	True
IE-CAE	Deep Learning	univariate	True
IF-LOF	Ourlier Detection	multivariate	True
KNN	Classic Machine Learning	multivariate	True
LOF	Ourlier Detection	multivariate	True
LSTM-AD	Deep Learning	multivariate	True
LSTM-VAE	Deep Learning	multivariate	True
LaserDBN	Stochastic Learning	multivariate	True
Left STAMPi	Data Mining	univariate	True
MSCRED	Deep Learning	multivariate	True
MTAD-GAT	Deep Learning	multivariate	True
MedianMethod	Statistics (Regression & Forecasting)	univariate	True
MultiHMM	Stochastic Learning	multivariate	True
NormA	Data Mining	univariate	True
Normalizing Flows	Deep Learning	multivariate	True
NoveltySVR	Classic Machine Learning	univariate	True
NumentaHTM	Deep Learning	univariate	True
OceanWNN	Deep Learning	univariate	True
OmniAnomaly	Deep Learning	multivariate	True
PCC	Classic Machine Learning	multivariate	True
PCI	Statistics (Regression & Forecasting)	univariate	True
PS-SVM	Classic Machine Learning	univariate	True
PST	Data Mining	univariate	True
RBForest	Classic Machine Learning	multivariate	True
RForest	Classic Machine Learning	univariate	True
RobustPCA	Classic Machine Learning	multivariate	True
S-H-ESD	Statistics (Regression & Forecasting)	univariate	True
SAND	Data Mining	univariate	True
SARIMA	Statistics (Regression & Forecasting)	univariate	True
SR	Signal Analysis	univariate	True
SR-CNN	Deep Learning	univariate	True
SSA	Data Mining	univariate	True
STAMP	Data Mining	univariate	True
STOMP	Data Mining	univariate	True
Series2Graph	Data Mining	univariate	True
Sub-Fast-MCD	Statistics (Regression & Forecasting)	univariate	True
Sub-IF	Ourlier Detection	univariate	True
Sub-LOF	Ourlier Detection	univariate	True
TARZAN	Data Mining	univariate	True
TAnoGAN	Deep Learning	multivariate	True
TSBitmap	Data Mining	univariate	True
Telemanom	Deep Learning	multivariate	True
Torsk	Deep Learning	multivariate	True
Triple ES	Statistics (Regression & Forecasting)	univariate	True
VALMOD	Data Mining	univariate	True
XGBoosting	Classic Machine Learning	univariate	True
iForest	Ourlier Detection	multivariate	True
k-Means	Classic Machine Learning	multivariate	True
AD-LTI	Deep Learning	multivariate	False
AMD Segmentation	Statistics (Regression & Forecasting)	univariate	False
ANODE	Statistics (Regression & Forecasting)		False
AOSVM	Classic Machine Learning	univariate	False
AR	Statistics (Regression & Forecasting)	univariate	False
ARMA	Statistics (Regression & Forecasting)	univariate	False
BLOF	Ourlier Detection		False
BoehmerGraph	Data Mining		False
Box Plot	Statistics (Regression & Forecasting)	univariate	False
CHEB	Statistics (Regression & Forecasting)	univariate	False
CoalESN	Deep Learning		False
ConInd	Statistics (Regression & Forecasting)	multivariate	False
CxDBN	Stochastic Learning	multivariate	False
DAD	Data Mining	univariate	False
DADS	Data Mining	univariate	False
DILOF	Ourlier Detection	multivariate	False
Deep K-Means	Deep Learning		False
Deep OCSVM	Deep Learning		False
DeepLSTM	Deep Learning	univariate	False
DeepPCA	Deep Learning		False
DissimilarityAlgo	Data Mining	univariate	False
Double ES (Holt's)	Statistics (Regression & Forecasting)		False
Dynamic State Estimator (DSE)	Statistics (Regression & Forecasting)		False
EDBN	Stochastic Learning		False
EM-HMM	Stochastic Learning		False
EWMA-STR	Statistics (Regression & Forecasting)		False
Eros-SVMs	Classic Machine Learning	multivariate	False
FuzzyDNBC	Stochastic Learning	multivariate	False
GLA	Stochastic Learning	univariate	False
GeckoFSM	Ourlier Detection	multivariate	False
GridLOF	Ourlier Detection		False
HMAD	Stochastic Learning	univariate	False
HSDE	Ourlier Detection	univariate	False
HSMM	Stochastic Learning	univariate	False
Hybrid K-means	Classic Machine Learning		False
I-HMM	Stochastic Learning	univariate	False
ILOF	Data Mining	multivariate	False
K-LOF	Ourlier Detection		False
KNN (PTSA)	Classic Machine Learning		False
Kalman Filter	Statistics (Regression & Forecasting)		False
KnorrSeq2	Data Mining		False
LAMP (GPU)	Deep Learning		False
LOCI/aLOCI	Ourlier Detection	multivariate	False
LSTM (PTSA)	Deep Learning		False
LSTM-based VAE-GAN	Deep Learning	multivariate	False
MA	Statistics (Regression & Forecasting)	univariate	False
MAD-GAN	Deep Learning	multivariate	False
MCD	Statistics (Regression & Forecasting)		False
MCOD	Ourlier Detection	univariate	False
MERLIN	Data Mining	univariate	False
MGDD	Statistics (Regression & Forecasting)	multivariate	False
MS-SVDD	Classic Machine Learning	multivariate	False
MoteESN	Deep Learning		False
MultiHTM	Deep Learning	multivariate	False
NetworkSVM	Classic Machine Learning	multivariate	False
NorM	Data Mining	univariate	False
NorM (SAD)	Data Mining		False
OC-KFD	Classic Machine Learning	multivariate	False
Online DWT-MLEAD	Signal Analysis	univariate	False
PAD	Deep Learning	univariate	False
PCA	Classic Machine Learning	multivariate	False
Poly (PTSA)	Statistics (Regression & Forecasting)	univariate	False
RADM	Deep Learning	multivariate	False
RPIF			False
RUSBoost	Classic Machine Learning	univariate	False
RePAD	Statistics (Regression & Forecasting)	univariate	False
Robust Deep AutoEncoder	Deep Learning		False
S-SVM	Classic Machine Learning	univariate	False
SALOF	Ourlier Detection		False
SCRIMP++	Data Mining		False
SH-ESD+	Statistics (Regression & Forecasting)	univariate	False
SLADE-MTS	Classic Machine Learning		False
SLADE-TS	Classic Machine Learning		False
STORN	Deep Learning	univariate	False
Simple ES (EWMA)	Statistics (Regression & Forecasting)		False
SmartSifter	Stochastic Learning	multivariate	False
Sparse AutoEncoder	Deep Learning		False
Structured Denoising AutoEncoder (StrDAE)	Deep Learning		False
SurpriseEncoding	Data Mining	multivariate	False
TCN-AE	Deep Learning	univariate	False
TOLF	Ourlier Detection		False
TwoFinger	Data Mining	univariate	False
U-GMM-HMM	Stochastic Learning	univariate	False
VELC	Deep Learning	univariate	False
Yesterday (PTSA)	Statistics (Regression & Forecasting)	univariate	False
pEWMA	Statistics (Regression & Forecasting)		False
sequenceMiner	Classic Machine Learning	univariate	False

Implementation Details

More than half of the 71 chosen algorithms had to be reimplemented by ourselves. However, some authors provided algorithm implementations or community versions exist. All implementations can be found in our Github repository.

Method Name	Source Code Origin	Language	License	Method Family
ARIMA	own (John Paparrizos and team)	Python	no license	forecasting	→Github
AE	own	Python, Tensorflow	MIT	reconstruction	→Github
Bagel	original	Python	no license	reconstruction	→Github
CBLOF	community (PyOD)	Python	BSD 2	distance	→Github
COF	community (PyOD)	Python	BSD 2	distance	→Github
COPOD	community (PyOD)	Python	BSD 2	distribution	→Github
DAE	own	Python, Tensorflow	MIT	reconstruction	→Github
DBStream	original	R	no license	distance	→Github
DeepAnT	own	Python, Pytorch	no license	forecasting	→Github
DeepNAP	own	Python, Pytorch	MIT	forecasting	→Github
Donut	original	Python, Pytorch	no license	reconstruction	→Github
DSPOT	original	Python	GPL 3.0	distribution	→Github
DWT-MLEAD	own	Python	MIT	distribution	→Github
EIF	original	Python	UIUC	trees	→Github
EncDec-AD	own	Python, Pytorch	MIT	reconstruction	→Github
Ensemble GI	own	Python	MIT	encoding	→Github
Fast-MCD	own	Python	MIT	distribution	→Github
FFT	own	Python	MIT	reconstruction	→Github
RForest	own	Python	MIT	forecasting	→Github
XGBoosting	own	Python	MIT	forecasting	→Github
GrammarViz	original	Java	GPL 2.0	encoding	→Github
HBOS	community (PyOD)	Python	BSD 2	distribution	→Github
HealthESN	own	Python	MIT	forecasting	→Github
HIF	original	Python	GPL 2.0	trees	→Github
HOT SAX	original	Python	GPL 2.0	distance	→Github
Hybrid KNN	own	Python, Pytorch	MIT	distance	→Github
IF-LOF	own	Python	MIT	trees	→Github
iForest	community (PyOD)	Python	BSD 2	trees	→Github
IE-CAE	own	Python, Pytorch	MIT	reconstruction	→Github
k-Means	own	Python	MIT	distance	→Github
KNN	community (PyOD)	Python	BSD 2	distance	→Github
LaserDBN	own	Python	MIT	encoding	→Github
Left STAMPi	original	Python	BSD	distance	→Github
LOF	community (PyOD)	Python	BSD 2	distance	→Github
LSTM-AD	own	Python, Pytorch	MIT	forecasting	→Github
LSTM-VAE	own	Python, Tensorflow	MIT	reconstruction	→Github
MedianMethod	own	Python	MIT	forecasting	→Github
MSCRED	own	Python, Tensorflow	MIT	reconstruction	→Github
MTAD-GAT	own	Python, Pytorch	MIT	forecasting	→Github
MultiHMM	own	Python	MIT	encoding	→Github
NormA	original	Python	private	distance	→Github
Normalizing Flows	own	Python, Pytorch	MIT	distribution	→Github
NoveltySVR	own	Python	GPL 3.0	forecasting	→Github
NumentaHTM	original	Python	AGPL	forecasting	→Github
OceanWNN	own	Python, Pytorch	MIT	forecasting	→Github
OmniAnomaly	original	Python, Tensorflow	MIT	reconstruction	→Github
PCC	community (PyOD)	Python	BSD 2	reconstruction	→Github
PCI	own	Python	MIT	reconstruction	→Github
PS-SVM	own	Python	MIT	distance	→Github
PST	own	R	GPL	encoding	→Github
RBForest	own	Python	MIT	forecasting	→Github
RobustPCA	community	Python	MIT	reconstruction	→Github
S-H-ESD	own	R	GPL 3.0	distribution	→Github
SAND	original	Python	private	distance	→Github
SARIMA	own	Python	BSD 3.0	forecasting	→Github
Series2Graph	original	Python	private	distance	→Github
SR	original	Python	MIT	reconstruction	→Github
SR-CNN	original	Python, Pytorch	MIT	reconstruction	→Github
SSA	own (John Paparrizos and team)	Python	no license	distance	→Github
STAMP	original	R	Apache	distance	→Github
STOMP	original	R	Apache	distance	→Github
Sub-Fast-MCD	own	Python	MIT	distribution	→Github
Sub-IF	own	Python	MIT	trees	→Github
Sub-LOF	own	Python	MIT	distance	→Github
TAnoGAN	own	Python, Pytorch	no license	reconstruction	→Github
TARZAN	original	Python	no license	encoding	→Github
Telemanom	original	Python, Tensorflow	Caltech	forecasting	→Github
Torsk	original	Python, Pytorch	no license	forecasting	→Github
Triple ES	own	Python	MIT	forecasting	→Github
TSBitmap	community	Python	no license	encoding	→Github
VALMOD	original	R	Apache	distance	→Github

Parameterization

After an independent parameter search, we conducted the experiments with the following parameters. Some parameters depend on data set properties.

ARIMA

Parameter	Value
window_size	1.0 dataset period size
max_lag	10% of dataset length
p_start	1
q_start	1
max_p	5
max_q	5
differencing_degree	1
distance_metric	twed
random_state	42

AE

Parameter	Value
latent_size	32
epochs	500
learning_rate	0.001
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
random_state	42

Bagel

Parameter	Value
window_size	2.0 dataset period size
latent_size	6
hidden_layer_shape	[100, 100]
dropout	0.1
cuda	False
epochs	500
batch_size	64
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
random_state	42

CBLOF

Parameter	Value
n_clusters	50
alpha	default
beta	default
use_weights	default
random_state	42
n_jobs	1

COF

Parameter	Value
n_neighbors	50
random_state	42

COPOD

Parameter	Value
random_state	42

DAE

Parameter	Value
latent_size	32
epochs	500
learning_rate	0.001
noise_ratio	0.1
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
random_state	42

DBStream

Parameter	Value
window_size	1.0 dataset period size
radius	1.3
lambda	0.001
distance_metric	euclidean
shared_density	True
n_clusters	30
alpha	0.5
min_weight	0
random_state	42

DeepAnT

Parameter	Value
epochs	500
window_size	0.5 dataset period size
prediction_window_size	50
learning_rate	0.001
batch_size	64
random_state	42
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10

DeepNAP

Parameter	Value
anomaly_window_size	max anomaly length
partial_sequence_length	3
lstm_layers	1
rnn_hidden_size	200
dropout	0.5
linear_hidden_size	100
batch_size	64
validation_batch_size	64
epochs	500
learning_rate	0.001
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
random_state	42

Donut

Parameter	Value
window_size	1.0 dataset period size
latent_size	5
regularization	0.001
linear_hidden_size	130
epochs	500
random_state	42

DSPOT

Parameter	Value
q	default
n_init	1000
level	0.99
up	True
down	True
alert	default
bounded	True
max_excess	200
random_state	42

DWT-MLEAD

Parameter	Value
start_level	3
quantile_epsilon	0.1
random_state	42

EIF

Parameter	Value
n_trees	500
max_samples	None
extension_level	None
limit	None
random_state	42

EncDec-AD

Parameter	Value
lstm_layers	3
anomaly_window_size	max anomaly length
latent_size	-30% default value
batch_size	64
validation_batch_size	64
epochs	500
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
learning_rate	0.001
random_state	42
window_size	1.0 dataset period size
test_batch_size	64

Ensemble GI

Parameter	Value
anomaly_window_size	max anomaly length
n_estimators	500
max_paa_transform_size	20
max_alphabet_size	10
selectivity	0.8
random_state	42
n_jobs	1
window_method	sliding

Fast-MCD

Parameter	Value
store_precision	True
support_fraction	default
random_state	42

FFT

Parameter	Value
fft_parameters	3
context_window_size	5
local_outlier_threshold	0.78
max_anomaly_window_size	max anomaly length
max_sign_change_distance	20
random_state	42

RForest

Parameter	Value
train_window_size	500
n_trees	500
max_features_method	auto
bootstrap	True
max_samples	None
random_state	42
verbose	0
n_jobs	1
max_depth	4
min_samples_split	2
min_samples_leaf	1

XGBoosting

Parameter	Value
train_window_size	500
n_estimators	500
learning_rate	0.001
booster	gbtree
tree_method	auto
n_trees	500
max_depth	4
max_samples	None
colsample_bytree	None
colsample_bylevel	None
colsample_bynode	None
random_state	42
verbose	0
n_jobs	1

GrammarViz

Parameter	Value
anomaly_window_size	max anomaly length
paa_transform_size	5
alphabet_size	6
normalization_threshold	0.01
random_state	42

HBOS

Parameter	Value
n_bins	20
alpha	default
bin_tol	default
random_state	42

HealthESN

Parameter	Value
linear_hidden_size	default
prediction_window_size	50
connectivity	default
spectral_radius	default
activation	default
random_state	42

HIF

Parameter	Value
n_trees	500
max_samples	None
random_state	42

HOT SAX

Parameter	Value
num_discords	None
anomaly_window_size	max anomaly length
paa_transform_size	3
alphabet_size	3
normalization_threshold	0.01
random_state	42

Hybrid KNN

Parameter	Value
linear_layer_shape	default
split	0.8
anomaly_window_size	max anomaly length
batch_size	64
test_batch_size	64
epochs	500
early_stopping_delta	0.05
early_stopping_patience	10
learning_rate	0.001
n_neighbors	10
n_estimators	500
random_state	42

IF-LOF

Parameter	Value
n_trees	500
max_samples	default
n_neighbors	50
alpha	default
m	default
random_state	42

iForest

Parameter	Value
n_trees	500
max_samples	None
max_features	1.0
bootstrap	false
random_state	42
verbose	0
n_jobs	1

IE-CAE

Parameter	Value
anomaly_window_size	max anomaly length
kernel_size	default
num_kernels	32
latent_size	100
leaky_relu_alpha	0.03
batch_size	64
test_batch_size	64
learning_rate	0.001
epochs	500
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
random_state	42

k-Means

Parameter	Value
n_clusters	50
anomaly_window_size	max anomaly length
stride	1
n_jobs	1
random_state	42

KNN

Parameter	Value
n_neighbors	50
leaf_size	20
method	default
radius	default
distance_metric_order	2
n_jobs	1
random_state	42

LaserDBN

Parameter	Value
timesteps	2
n_bins	10
random_state	42

Left STAMPi

Parameter	Value
anomaly_window_size	max anomaly length
n_init_train	10% of dataset length or until first anomaly
random_state	42

LOF

Parameter	Value
n_neighbors	50
leaf_size	20
distance_metric_order	2
n_jobs	1
random_state	42

LSTM-AD

Parameter	Value
lstm_layers	1
split	0.8
window_size	2.0 dataset period size
prediction_window_size	50
batch_size	64
validation_batch_size	64
epochs	500
early_stopping_delta	0.05
early_stopping_patience	10
learning_rate	0.001
random_state	42
test_batch_size	64

LSTM-VAE

Parameter	Value
rnn_hidden_size	5
latent_size	5
learning_rate	0.001
batch_size	64
epochs	500
window_size	1.0 dataset period size
lstm_layers	10
early_stopping_delta	0.05
early_stopping_patience	10

MedianMethod

Parameter	Value
neighbourhood_size	2.0 dataset period size
random_state	42

MSCRED

Parameter	Value
windows	default
gap_time	10
window_size	1.0 dataset period size
batch_size	64
learning_rate	0.001
epochs	500
early_stopping_patience	10
early_stopping_delta	0.05
split	0.8
test_batch_size	64
random_state	42

MTAD-GAT

Parameter	Value
mag_window_size	40
score_window_size	52
threshold	6
context_window_size	30
kernel_size	7
learning_rate	0.001
epochs	500
batch_size	64
window_size	2.0 dataset period size
gamma	0.8
latent_size	5
linear_layer_shape	[5, 5, 5]
early_stopping_patience	10
early_stopping_delta	0.05
split	0.8
random_state	42

MultiHMM

Parameter	Value
discretizer	choquet
n_bins	5
random_state	42

NormA

Parameter	Value
anomaly_window_size	max anomaly length
normal_model_percentage	0.5
random_state	42

Normalizing Flows

Parameter	Value
n_hidden_features_factor	1.0
hidden_layer_shape	[100, 100]
window_size	1.0 dataset period size
split	0.8
epochs	500
batch_size	64
test_batch_size	64
teacher_epochs	100
distillation_iterations	100
percentile	0.05
early_stopping_patience	10
early_stopping_delta	0.05
random_state	42

NoveltySVR

Parameter	Value
n_init_train	10% of dataset length or until first anomaly
forgetting_time	None
train_window_size	500
anomaly_window_size	max anomaly length
lower_suprise_bound	None
scaling	standard
epsilon	0.1
verbose	0
C	1.0
kernel	rbf
degree	3
gamma	None
coef0	0.0
tol	0.001
stabilized	True
random_state	42

NumentaHTM

Parameter	Value
encoding_input_width	21
encoding_output_width	75
autoDetectWaitRecords	50
columnCount	2048
numActiveColumnsPerInhArea	50
potentialPct	0.1
synPermConnected	0.1
synPermActiveInc	0.05
synPermInactiveDec	0.01
cellsPerColumn	32
inputWidth	2048
newSynapseCount	15
maxSynapsesPerSegment	32
maxSegmentsPerCell	128
initialPerm	0.15
permanenceInc	0.1
permanenceDec	0.1
globalDecay	0
maxAge	0
minThreshold	9
activationThreshold	12
pamLength	1
alpha	0.5
random_state	42

OceanWNN

Parameter	Value
train_window_size	500
hidden_size	20
batch_size	64
test_batch_size	64
epochs	500
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10
learning_rate	0.001
wavelet_a	-3.25
wavelet_k	-1.95
wavelet_wbf	mexican_hat
wavelet_cs_C	2.275
threshold_percentile	0.99
random_state	42
with_threshold	True

OmniAnomaly

Parameter	Value
latent_size	4
rnn_hidden_size	100
window_size	1.0 dataset period size
linear_hidden_size	100
nf_layers	5
epochs	500
split	0.8
batch_size	64
l2_reg	0.0001
learning_rate	0.001
random_state	42

PCC

Parameter	Value
n_components	default
n_selected_components	default
whiten	default
svd_solver	auto
tol	default
max_iter	default
random_state	42

PCI

Parameter	Value
window_size	0.5 dataset period size
thresholding_p	0.05
random_state	42

PS-SVM

Parameter	Value
embed_dim_range	[0.5, 1.0, 1.5] * dataset period size
project_phasespace	False
nu	0.5
kernel	rbf
gamma	None
degree	3
coef0	0.0
tol	0.001
random_state	42

PST

Parameter	Value
window_size	1.0 dataset period size
max_depth	4
n_min	1
y_min	default
n_bins	5
sim	SIMn
random_state	42

RBForest

Parameter	Value
train_window_size	500
n_estimators	500
max_features_per_estimator	0.5
n_trees	500
max_features_method	auto
bootstrap	True
max_samples	None
random_state	42
verbose	0
n_jobs	1
max_depth	4
min_samples_split	2
min_samples_leaf	1

RobustPCA

Parameter	Value
max_iter	default
random_state	42

S-H-ESD

Parameter	Value
max_anomalies	dataset contamination
timestamp_unit	m
random_state	42

SAND

Parameter	Value
anomaly_window_size	max anomaly length
n_clusters	50
n_init_train	10% of dataset length or until first anomaly
iter_batch_size	500
alpha	0.5
random_state	42

SARIMA

Parameter	Value
train_window_size	500
prediction_window_size	50
max_lag	10% of dataset length
period	dataset period size
max_iter	default
exhaustive_search	False
n_jobs	1
random_state	42

Series2Graph

Parameter	Value
window_size	1.0 dataset period size
query_window_size	1.5*window_size
rate	100
random_state	42

SR

Parameter	Value
mag_window_size	40
score_window_size	40
window_size	1.0 dataset period size
random_state	42

SR-CNN

Parameter	Value
window_size	1.5 dataset period size
random_state	42
step	64
num	10
learning_rate	0.001
epochs	500
batch_size	64
n_jobs	1
split	0.8
early_stopping_delta	0.05
early_stopping_patience	10

SSA

Parameter	Value
ep	3
window_size	2.0 dataset period size
rf_method	alpha
alpha	0.2
random_state	42

STAMP

Parameter	Value
anomaly_window_size	max anomaly length
exclusion_zone	0.5
verbose	0
n_jobs	1
random_state	42

STOMP

Parameter	Value
anomaly_window_size	max anomaly length
exclusion_zone	0.5
verbose	0
n_jobs	1
random_state	42

Sub-Fast-MCD

Parameter	Value
store_precision	True
support_fraction	default
random_state	42

Sub-IF

Parameter	Value
window_size	1.0 dataset period size
n_trees	500
max_samples	None
max_features	1.0
bootstrap	false
random_state	42
verbose	0
n_jobs	1

Sub-LOF

Parameter	Value
window_size	1.0 dataset period size
n_neighbors	50
leaf_size	20
distance_metric_order	2
n_jobs	1
random_state	42

TAnoGAN

Parameter	Value
epochs	500
cuda	False
window_size	1.0 dataset period size
learning_rate	0.001
batch_size	64
n_jobs	1
random_state	42
early_stopping_patience	10
early_stopping_delta	0.05
split	0.8
iterations	25

TARZAN

Parameter	Value
random_state	42
anomaly_window_size	max anomaly length
alphabet_size	4

Telemanom

Parameter	Value
batch_size	64
smoothing_window_size	30
smoothing_perc	0.05
error_buffer	100
dropout	0.5
lstm_batch_size	64
epochs	500
split	0.8
early_stopping_patience	10
early_stopping_delta	0.05
window_size	1.5 dataset period size
prediction_window_size	50
p	0.17
random_state	42

Torsk

Parameter	Value
input_map_size	100
input_map_scale	0.125
context_window_size	10
train_window_size	100
prediction_window_size	5
transient_window_size	20% of train_window_size
spectral_radius	2.0
density	0.01
reservoir_representation	sparse
imed_loss	False
train_method	pinv_svd
tikhonov_beta	None
verbose	0
scoring_small_window_size	10
scoring_large_window_size	100
random_state	42

Triple ES

Parameter	Value
train_window_size	500
period	dataset period size
trend	add
seasonal	add
random_state	42

TSBitmap

Parameter	Value
feature_window_size	500
lead_window_size	200
lag_window_size	500
alphabet_size	4
level_size	2
compression_ratio	1
random_state	42

VALMOD

Parameter	Value
min_anomaly_window_size	1.0 dataset period size
max_anomaly_window_size	2.0 dataset period size
heap_size	50
exclusion_zone	0.5
verbose	0
random_state	42