ELKI command line parameter overview:

Kernel function to use.

Number of top outliers to compute.

Parameter for:

Frequency of zeros in the projection matrix.

Default: 3.0

Parameter for:

Similarity function to derive the distance between database objects from.

Algorithm to run.

Distance function to determine the distance between database objects.

Number of stable iterations for convergence.

Default: 15

Parameter for:

Distance function to use.

Similarity matrix initialization..

Dampening factor lambda. Usually 0.5 to 1.

Default: 0.5

Parameter for:

Maximum number of iterations.

Default: 1000

Parameter for:

Quantile to use for diagonal entries.

Default: 0.5

Parameter for:

Similarity function to use.

Pattern to recognize class label attributes.

Default: (Class|Class-?Label)

Parameter for:

Pattern to recognize external ID attributes.

Default: (External-?ID)

Parameter for:

Include the query object in the evaluation.

Default: false

Parameter for:

K to compute the average precision at.

Parameter for:

Relative amount of object to sample.

Parameter for:

Random seed for deterministic sampling.

Default: use global random seed

Parameter for:

Scale the data space extension by the given factor.

Default: 1.0

Parameter for:

Subspace dimensionality to search for.

Parameter for:

Population size for evolutionary algorithm.

Parameter for:

The number of equi-depth grid ranges to use in each dimension.

Parameter for:

The random number generator seed.

Default: use global random seed

Parameter for:

Alpha parameter to control the shape of the output distribution.

Default: 0.1

Parameter for:

A list of the distribution estimators to try.

KMeans variant

Half-transparent filling of bubbles.

Default: false

Parameter for:

Additional scaling function for bubbles.

Bundle file to load the data from.

Parameter for:

Flag to indicate that only subspaces with large coverage (i.e. the fraction of the database that is covered by the dense units) are selected, the rest will be pruned.

Default: false

Parameter for:

Pattern to recognize noise classes by their label.

Parameter for:

Pattern to recognize noise models by their label.

Parameter for:

The random generator seed.

Default: use global random seed

Parameter for:

Pattern to specify clusters to reassign.

Parameter for:

Factor for scaling the specified cluster sizes.

Default: 1.0

Parameter for:

The generator specification file.

Parameter for:

Distance function to use for computing the c-index.

Control how noise should be treated.

Default: TREAT_NOISE_AS_SINGLETONS

Parameter for:

Inclusion threshold for canopy clustering. t1 >= t2!

Parameter for:

Removal threshold for canopy clustering. t1 >= t2!

Parameter for:

Flag to indicate that an adjustment of the applied heuristic for choosing an interval is performed after an interval is selected.

Default: false

Parameter for:

The maximum jitter for distance values.

Parameter for:

The maximum level for splitting the hypercube.

Parameter for:

The minimum dimensionality of the subspaces to be found.

Default: 1

Parameter for:

Threshold for minimum number of points in a cluster.

Parameter for:

Parameter for multiple node deletion to accelerate the algorithm.

Default: 1.0

Parameter for:

Threshold value to determine the maximal acceptable score (mean squared residue) of a bicluster.

Parameter for:

The number of biclusters to be found.

Default: 1

Parameter for:

Distribution of replacement values when masking found clusters.

Random generator seed.

Default: use global random seed

Parameter for:

Number of samples (iterations) to run.

Default: 5

Parameter for:

The size of the sample.

Parameter for:

Class label to use for negative instances.

Default: negative

Parameter for:

Regular expression to identify positive objects.

Parameter for:

Class label to use for positive instances.

Default: positive

Parameter for:

Maximum value to allow.

Parameter for:

Minimum value to allow.

Parameter for:

Flag to indicate that only subspaces with large coverage (i.e. the fraction of the database that is covered by the dense units) are selected, the rest will be pruned.

Default: false

Parameter for:

The density threshold for the selectivity of a unit, where the selectivity isthe fraction of total feature vectors contained in this unit.

Parameter for:

The number of intervals (units) in each dimension.

Parameter for:

Parameter to override the clustering label, mostly to give a more descriptive label.

Parameter for:

Output file name. When not given, the result will be written to stdout.

Parameter for:

Always append to the output file.

Default: false

Parameter for:

The number of neighbors (not including the query object) to use for computing the COF score.

Parameter for:

number of bins

Default: 50

Parameter for:

Class label for the 'positive' class.

Parameter for:

Class label for the 'positive' class.

Parameter for:

Class to use as scaling function.

Class to use as scaling function.

Control how noise should be treated.

Default: TREAT_NOISE_AS_SINGLETONS

Parameter for:

Distance function to use for measuring concordant and discordant pairs.

The assumed distribution of squared distances. ChiSquared is faster, Gamma expected to be more accurate but could also overfit.

Default: GAMMA

Parameter for:

Expected share of outliers. Only affect score normalization.

Default: 0.001

Parameter for:

The number of nearest neighbors of an object to be considered for computing its COP_SCORE.

Parameter for:

The number of nearest neighbors of an object to be considered for computing its COP_SCORE.

Parameter for:

Include COP models (error vectors) in output. This needs more memory.

Default: false

Parameter for:

The class to compute (filtered) PCA.

The class to compute (filtered) PCA.

Number of neighbors to use for PCA.

Parameter for:

Phi parameter, expected rate of outliers. Set to 0 to use raw CDF values.

Parameter for:

Distance function to determine the distance between objects.

Expansion rate of the tree (Default: 1.3).

Default: 1.3

Parameter for:

Truncate tree when branches have less than this number of instances.

Default: 10

Parameter for:

Distance function to use for computing the davies-bouldin index.

Control how noise should be treated.

Default: TREAT_NOISE_AS_SINGLETONS

Parameter for:

Database class.

Database indexes to add.

Database connection class.

Class label class to use.

The index of the label to be used as class label. The first label is 0, negative indexes are relative to the end.

Parameter for:

Dimensionality of the vectors to generate.

Parameter for:

The index of the label to be used as external Id. The first label is 0; negative indexes are relative to the end.

Parameter for:

The filters to apply to the input data.

Seed for randomly generating vectors

Default: use global random seed

Parameter for:

The name of the input file to be parsed.

Parameter for:

Parser to provide the database.

Database size to generate.

Parameter for:

Object ID to start counting with

Default: 0

Parameter for:

size of the D-neighborhood

Parameter for:

minimum fraction of objects that must be outside the D-neighborhood of an outlier

Parameter for:

The maximum radius of the neighborhood to be considered.

Parameter for:

Threshold for minimum number of points in the epsilon-neighborhood of a point. The suggested value is '2 * dim - 1'.

Parameter for:

Threshold for minimum number of points within a cluster.

Parameter for:

Drawing style for dendrograms.

Default: RECTANGULAR

Parameter for:

Threshold for output accuracy fraction digits.

Default: 4

Parameter for:

Flag to use random sample (use knn query around centroid, if flag is not set).

Default: false

Parameter for:

Threshold for the size of the random sample to use. Default value is size of the complete dataset.

Parameter for:

an integer between 1 and the dimensionality of the feature space 1 specifying the dimension to be considered for distance computation.

Parameter for:

The maximum radius of the neighborhood to be considered in each dimension for determination of the preference vector.

Default: 0.001

Parameter for:

A comma separated list of positive doubles specifying the maximum radius of the neighborhood to be considered in each dimension for determination of the preference vector (default is 0.001 in each dimension). If only one value is specified, this value will be used for each dimension.

Default: 0.001

Parameter for:

Positive threshold for minumum numbers of points in the epsilon-neighborhood of a point. The value of the preference vector in dimension d_i is set to 1 if the epsilon neighborhood contains more than dish.minpts points and the following condition holds: for all dimensions d_j: |neighbors(d_i) intersection neighbors(d_j)| >= dish.minpts.

Parameter for:

The minimum number of points as a smoothing factor to avoid the single-link-effekt.

Default: 1

Parameter for:

The strategy for determination of the preference vector, available strategies are: [APRIORI| MAX_INTERSECTION](default is MAX_INTERSECTION)

Default: MAX_INTERSECTION

Parameter for:

a comma separated array of integer values, where 0 <= d_i < the dimensionality of the feature space specifying the dimensions to be considered for distance computation. If this parameter is not set, no dimensions will be considered, i.e. the distance between two objects is always 0.

Parameter for:

The dimension containing the latitude.

Parameter for:

The dimension containing the longitude.

Parameter for:

The name of the file containing the distance matrix.

Parameter for:

The name of the file containing the distance matrix.

Parameter for:

Parser used to load the distance matrix.

Weights to use for the distance function.

Parameter for:

Distance index to use.

Beta distribution alpha parameter

Parameter for:

Beta distribution beta parameter

Parameter for:

Cauchy distribution gamma/shape parameter.

Parameter for:

Chi distribution degrees of freedom parameter.

Parameter for:

Constant value.

Parameter for:

Exponential distribution rate (lambda) parameter (inverse of scale).

Parameter for:

Gamma distribution k = alpha parameter.

Parameter for:

Gamma distribution theta = 1/beta parameter.

Parameter for:

First shape parameter of kappa distribution.

Parameter for:

Second shape parameter of kappa distribution.

Parameter for:

Laplace distribution rate (lambda) parameter (inverse of scale).

Parameter for:

Distribution location parameter

Parameter for:

Shift offset parameter.

Parameter for:

Shift offset parameter.

Parameter for:

Mean of the distribution before logscaling.

Parameter for:

Standard deviation of the distribution before logscaling.

Parameter for:

Shifting offset, so the distribution does not begin at 0.

Default: 0.0

Parameter for:

Maximum value of distribution.

Parameter for:

Minimum value of distribution.

Parameter for:

Number of trials.

Parameter for:

Success probability.

Parameter for:

Random generation data source.

Default: use global random seed

Parameter for:

Distribution scale parameter

Parameter for:

Distribution shape parameter

Parameter for:

Skew of the distribution.

Parameter for:

Degrees of freedom.

Parameter for:

Ignore zero distances, beneficial for data sets with many duplicates.

Default: false

Parameter for:

Quantile to compute.

Default: 0.1

Parameter for:

Number of distances to compute, either relative (values less than 1), or absolute.

Parameter for:

Random generator seed.

Default: use global random seed

Parameter for:

Number of bins to use in the histogram. By default, it is only guaranteed to be within 1*n and 2*n of the given number.

Default: 20

Parameter for:

In a first pass, compute the exact minimum and maximum, at the cost of O(2*n*n) instead of O(n*n). The number of resulting bins is guaranteed to be as requested.

Default: false

Parameter for:

Enable sampling of O(n) size to determine the minimum and maximum distances approximately. The resulting number of bins can be larger than the given n.

Default: false

Parameter for:

Minimum relative density for a set of points to be considered a cluster (|C|>=doc.alpha*|S|).

Default: 0.2

Parameter for:

Preference of cluster size versus number of relevant dimensions (higher value means higher priority on larger clusters).

Default: 0.8

Parameter for:

Use heuristics as described, thus using the FastDOC algorithm (not yet implemented).

Default: false

Parameter for:

Random seed, for reproducible experiments.

Default: use global random seed

Parameter for:

Maximum extent of scattering of points along a single attribute for the attribute to be considered relevant.

Default: 0.05

Parameter for:

Radius increase factor.

Default: 1.1

Parameter for:

Number of neighbors to get for DWOF score outlier detection.

Parameter for:

The band size for time series alignment. By default, no constraint is used. If the value is larger than 0, it will be considered absolute, otherwise relative to the longer sequence. Note that 0 does not make sense: use Euclidean distance then instead.

Parameter for:

the delta parameter (similarity threshold) for EDR (positive number)

Default: 1.0

Parameter for:

Method to choose the initial cluster centers.

The termination criterion for maximization of E(M): E(M) - E(M') < em.delta

Default: 1.0E-7

Parameter for:

The number of clusters to find.

Parameter for:

Model factory.

Parameter to enable debugging for particular packages.

Parameter for:

Quantile to use in median voting.

Default: 0.5

Parameter for:

Voting strategy to use in the ensemble.

Number of neighbors to use for PCA.

Parameter for:

Threshold for approximate linear dependency: the strong eigenvectors of q are approximately linear dependent from the strong eigenvectors p if the following condition holds for all stroneg eigenvectors q_i of q (lambda_q < lambda_p): q_i' * M^check_p * q_i <= delta^2.

Default: 0.1

Parameter for:

Threshold for the maximum distance between two approximately linear dependent subspaces of two objects p and q (lambda_q < lambda_p) before considering them as parallel.

Default: 0.1

Parameter for:

The g parameter of ERP - comparison value to use in gaps.

Default: 0.0

Parameter for:

Class to evaluate the results with.

The inner neighborhood predicate to use.

The inner neighborhood predicate to use.

The number of steps allowed in the neighborhood graph.

Parameter for:

The number of steps allowed in the neighborhood graph.

Parameter for:

Filename with the precomputed k nearest neighbors.

Parameter for:

The file name containing the (external) cluster vector.

Parameter for:

The file listing the neighbors.

Parameter for:

The file name containing the (external) outlier scores.

Parameter for:

The pattern to match object ID prefix

Default: ^ID=

Parameter for:

Flag to signal an inverted outlier score.

Default: false

Parameter for:

Class to use as scaling function.

The pattern to match object score prefix

Parameter for:

Keep the first object chosen (which is chosen randomly) for the farthest points heuristic.

Default: false

Parameter for:

Number of nearest neighbors to use for ABOD.

Parameter for:

Random seed for generating projections.

Default: use global random seed

Parameter for:

Use the breadth first combinations instead of the cumulative sum approach

Default: false

Parameter for:

The number of instances to use in the ensemble.

Parameter for:

Specify a particular random seed.

Default: use global random seed

Parameter for:

The number of samples to draw from each uncertain object to determine the epsilon-neighborhood.

Parameter for:

Random generator used to draw samples.

Default: use global random seed

Parameter for:

The amount of samples that have to be epsilon-close for two objects to be neighbors.

Default: 0.5

Parameter for:

Dimensionality of vectors to retain.

Parameter for:

Regularize scores before using Gamma scaling.

Default: false

Parameter for:

Invert the value range to [0:1], with 1 being outliers instead of 0.

Default: false

Parameter for:

Core point predicate for Generalized DBSCAN

Use a model that keeps track of core points. Needs more memory.

Default: false

Parameter for:

Neighborhood predicate for Generalized DBSCAN

The number of reference points to be generated.

Parameter for:

Random generator seed.

Default: use global random seed

Parameter for:

Scale the grid by the given factor. This can be used to obtain reference points outside the used data space.

Default: 1.0

Parameter for:

Earth model to use for projection. Default: spherical model.

Filter to use for dimensionality reduction.

Significance niveau

Parameter for:

k nearest neighbors to use

Parameter for:

Scale the grid by the given factor. This can be used to obtain reference points outside the used data space.

Default: 1.0

Parameter for:

The number of partitions in each dimension. Points will be placed on the edges of the grid, except for a grid size of 0, where only the mean is generated as reference point.

Default: 1

Parameter for:

Width of the grid used, must be at least two times epsilon.

Parameter for:

Produce a hierarchical output.

Default: false

Parameter for:

The minimum cluster size.

Default: 1

Parameter for:

The minimum cluster size.

Default: 1

Parameter for:

Threshold for minimum number of points in the epsilon-neighborhood of a point (including this point).

Parameter for:

The threshold for 'strong' eigenvectors: the 'strong' eigenvectors explain a portion of at least alpha of the total variance.

Default: 0.85

Parameter for:

Threshold of a distance between a vector q and a given space that indicates that q adds a new dimension to the space.

Default: 0.25

Parameter for:

Optional parameter to specify the number of nearest neighbors considered in the PCA. If this parameter is not set, k is set to the value of parameter mu.

Parameter for:

Specifies the smoothing factor. The mu-nearest neighbor is used to compute the correlation reachability of an object.

Parameter for:

The Algorithm that performs the actual outlier detection on the resulting set of subspace

The discriminance value that determines the size of the test statistic .

Default: 0.1

Parameter for:

The threshold that determines how many d-dimensional subspace candidates to retain in each step of the generation

Default: 100

Parameter for:

The number of iterations in the Monte-Carlo processing.

Default: 50

Parameter for:

The random seed.

Default: use global random seed

Parameter for:

The statistical test that is used to calculate the deviation of two data samples

Generate a truncated hierarchical clustering result (or strict partitions).

Default: false

Parameter for:

Linkage method to use (e.g. Ward, Single-Link)

Merge singleton clusters into parent. This produces a more complex hierarchy, but that is easier to understand.

Default: false

Parameter for:

The minimum number of clusters to extract (there may be more clusters when tied).

Parameter for:

The threshold level for which to extract the clusters.

Parameter for:

The thresholding mode to use for extracting clusters: by desired number of clusters, or by distance threshold.

Default: BY_MINCLUSTERS

Parameter for:

Max. Hilbert-Level

Default: 32

Parameter for:

Compute up to k next neighbors

Default: 5

Parameter for:

Compute n outliers

Default: 10

Parameter for:

output of Top n or all elements

Default: TopN

Parameter for:

The maximum absolute variance along a coordinate axis.

Default: 0.01

Parameter for:

The number of nearest neighbors considered to determine the preference vector. If this value is not defined, k ist set to three times of the dimensionality of the database objects.

Parameter for:

Use separate frequencies for outliers and non-outliers.

Default: false

Parameter for:

Random generator seed for holdout evaluation.

Default: use global random seed

Parameter for:

Nearest neighbor to use for the statistic

Default: 1

Parameter for:

Maximum values in each dimension. If no value is specified, the maximum value in each dimension will be used. If only one value is specified, this value will be used for all dimensions.

Parameter for:

Minimum values in each dimension. If no value is specified, the minimum value in each dimension will be used. If only one value is specified, this value will be used for all dimensions.

Parameter for:

The number of times to repeat the experiment (default: 1)

Default: 1

Parameter for:

Number of object / random samples to analyze.

Parameter for:

The random number generator.

Default: use global random seed

Parameter for:

The dimensionality of the histogram in hue, saturation and brightness.

Parameter for:

Alpha value for hull drawing (in projected space!).

Default: Infinity

Parameter for:

Class to estimate ID from distance distribution.

Number of nearest neighbors to use for ID estimation (usually 20-100).

Parameter for:

Number of DBID to generate.

Parameter for:

First integer DBID to generate.

Default: 0

Parameter for:

Estimation method for intrinsic dimensionality.

Number of kNN (absolute or relative)

Default: 50.0

Parameter for:

Sample size (absolute or relative)

Default: 0.1

Parameter for:

Distance function to build the index for.

Number of reference points to use.

Parameter for:

Method to choose the reference points.

Estimator of intrinsic dimensionality.

Context set size (ID estimation).

Parameter for:

Reference set size.

Parameter for:

Partially transparent filling of index pages.

Default: false

Parameter for:

The pagefile factory for storing the index.

The number of nearest neighbors of an object to be considered for computing its INFLO score.

Parameter for:

The pruning threshold

Default: 1.0

Parameter for:

Maximum length of frequent itemsets to report. This can help to reduce the output size to only the most interesting patterns.

Parameter for:

Minimum length of frequent itemsets to report. This can help to reduce the output size to only the most interesting patterns.

Parameter for:

Threshold for minimum support as minimally required number of transactions (if > 1) or the minimum frequency (if <= 1).

Parameter for:

Jitter amount relative to data.

Parameter for:

Jitter random seed.

Default: use global random seed

Parameter for:

The data sources to join.

The data sources to join.

The data sources to join.

Port for the JSON web server to listen on.

Default: 8080

Parameter for:

Maximum leaf size for the k-d-tree. Nodes will be split until their size is smaller than this threshold.

Default: 1

Parameter for:

Intrinsic dimensionality of this data set. Use -1 for using the true data dimensionality, but values such as 0-2 often offer better performance.

Default: -1

Parameter for:

Maximum value of k to analyze.

Parameter for:

Minimum value of k to analyze.

Parameter for:

Kernel density function to use.

Minimum bandwidth for kernel density estimation.

Parameter for:

Scaling factor for the kernel function.

Default: 0.5

Parameter for:

Standard deviation of the laplace RBF kernel.

Default: 1.0

Parameter for:

The bias of the polynomial kernel, a constant that is added to the scalar product.

Parameter for:

The degree of the polynomial kernel function. Default: 2

Default: 2

Parameter for:

Constant term in the rational quadratic kernel.

Default: 1.0

Parameter for:

Standard deviation of the Gaussian RBF kernel.

Default: 1.0

Parameter for:

Sigmoid c parameter (scaling).

Default: 1.0

Parameter for:

Sigmoid theta parameter (bias).

Default: 0.0

Parameter for:

Dimension to use for clustering. For one-dimensional data, use 0.

Default: 0

Parameter for:

Kernel function for density estimation.

Number of nearest neighbors to use for bandwidth estimation.

Parameter for:

Kernel density estimation mode (baloon estimator vs. sample point estimator).

Default: BALLOON

Parameter for:

Half width of sliding window to find local minima.

Parameter for:

Kernel to use for kernel density LOF.

Clustering algorithm to use for detecting outliers.

KMeans variant to run multiple times.

KMeans variant to run multiple times.

Number of blocks to use for processing. Means will be recomputed after each block.

Default: 10

Parameter for:

Random source for producing blocks.

Default: use global random seed

Parameter for:

Method to choose the initial means.

The number of clusters to find.

Parameter for:

The maximum number of iterations to do. 0 means no limit.

Parameter for:

Initial means for k-means.

Parameter for:

Quality measure variant for deciding which run to keep.

Sample set size (if > 1) or sampling rante (if < 1).

Parameter for:

The random number generator seed.

Default: use global random seed

Parameter for:

The number of trials to run.

Parameter for:

Compute the final clustering variance statistic. Needs an additional full pass over the data set.

Default: false

Parameter for:

Compute the final clustering variance statistic. Needs an additional full pass over the data set.

Default: false

Parameter for:

Automatically open the result file.

Default: false

Parameter for:

Use simpler KML objects, compatibility mode.

Default: false

Parameter for:

Additional scaling function for KML colorization.

Number of neighbors to retreive for kNN benchmarking.

Parameter for:

Data source for the queries. If not set, the queries are taken from the database.

Random generator for sampling.

Default: use global random seed

Parameter for:

Sampling size parameter. If the value is less or equal 1, it is assumed to be the relative share. Larger values will be interpreted as integer sizes. By default, all data will be used.

Parameter for:

The number of neighbors to take into account for classification.

Default: 1

Parameter for:

Specifies the distance of the k-distant object to be assessed, ignoring the query object.

Parameter for:

The percentage of objects to use for sampling, or the absolute number of samples.

Default: 1.0

Parameter for:

Random generator seed for sampling.

Default: use global random seed

Parameter for: