ELKI

ELKI: Environment for DeveLoping KDD-Applications Supported by Index-Structures.

See:
Description

Algorithms

de.lmu.ifi.dbs.elki.algorithm	Algorithms suitable as a task for the KDDTask main routine.
de.lmu.ifi.dbs.elki.algorithm.clustering	Clustering algorithms Clustering algorithms are supposed to implement the Algorithm-Interface.
de.lmu.ifi.dbs.elki.algorithm.clustering.correlation	Correlation clustering algorithms
de.lmu.ifi.dbs.elki.algorithm.clustering.correlation.cash	Helper classes for the CASH algorithm.
de.lmu.ifi.dbs.elki.algorithm.clustering.subspace	Axis-parallel subspace clustering algorithms The clustering algorithms in this package are instances of both, projected clustering algorithms or subspace clustering algorithms according to the classical but somewhat obsolete classification schema of clustering algorithms for axis-parallel subspaces.
de.lmu.ifi.dbs.elki.algorithm.clustering.subspace.clique	Helper classes for the CLIQUE algorithm.
de.lmu.ifi.dbs.elki.algorithm.clustering.trivial	Trivial clustering algorithms: all in one, no clusters, label clusterings These methods are mostly useful for providing a reference result in evaluation.
de.lmu.ifi.dbs.elki.algorithm.outlier	Outlier detection algorithms
de.lmu.ifi.dbs.elki.algorithm.outlier.meta	Meta outlier detection algorithms: external scores, score rescaling.
de.lmu.ifi.dbs.elki.algorithm.outlier.spatial	Spatial outlier detection algorithms
de.lmu.ifi.dbs.elki.algorithm.outlier.spatial.neighborhood	Spatial outlier neighborhood classes
de.lmu.ifi.dbs.elki.algorithm.outlier.spatial.neighborhood.weighted	Weighted Neighborhood definitions.
de.lmu.ifi.dbs.elki.algorithm.outlier.trivial	Trivial outlier detection algorithms: no outliers, all outliers, label outliers.
de.lmu.ifi.dbs.elki.algorithm.statistics	Statistical analysis algorithms The algorithms in this package perform statistical analysis of the data (e.g. compute distributions, distance distributions etc.)

Databases and Index Structures

de.lmu.ifi.dbs.elki.database	ELKI database layer - loading, storing, indexing and accessing data
de.lmu.ifi.dbs.elki.database.datastore	General data store layer API (along the lines of Map<DBID, T> - use everywhere!)
de.lmu.ifi.dbs.elki.database.datastore.memory	Memory data store implementation for ELKI.
de.lmu.ifi.dbs.elki.database.ids	Database object identification and ID group handling API.
de.lmu.ifi.dbs.elki.database.ids.generic	Database object identification and ID group handling - generic implementations.
de.lmu.ifi.dbs.elki.database.ids.integer	Integer-based DBID implementation -- do not use directly - always use DBIDUtil.
de.lmu.ifi.dbs.elki.database.query	Database queries - computing distances, neighbors, similarities - API and general documentation.
de.lmu.ifi.dbs.elki.database.query.distance	Prepared queries for distances.
de.lmu.ifi.dbs.elki.database.query.knn	Prepared queries for k nearest neighbor (kNN) queries.
de.lmu.ifi.dbs.elki.database.query.range	Prepared queries for ε-range queries.
de.lmu.ifi.dbs.elki.database.query.rknn	Prepared queries for reverse k nearest neighbor (rkNN) queries.
de.lmu.ifi.dbs.elki.database.query.similarity	Prepared queries for similarity functions.
de.lmu.ifi.dbs.elki.database.relation	Relations, materialized and virtual (views).
de.lmu.ifi.dbs.elki.datasource	Data normalization (and reconstitution) of data sets.
de.lmu.ifi.dbs.elki.datasource.bundle	Object bundles - exchange container for multi-represented objects.
de.lmu.ifi.dbs.elki.datasource.filter	Data filtering, in particular for normalization and projection.
de.lmu.ifi.dbs.elki.datasource.parser	Parsers for different file formats and data types.
de.lmu.ifi.dbs.elki.index	Index structure implementations
de.lmu.ifi.dbs.elki.index.preprocessed	Index structure based on preprocessors
de.lmu.ifi.dbs.elki.index.preprocessed.knn	Indexes providing KNN and rKNN data.
de.lmu.ifi.dbs.elki.index.preprocessed.localpca	Index using a preprocessed local PCA.
de.lmu.ifi.dbs.elki.index.preprocessed.preference	Indexes storing preference vectors.
de.lmu.ifi.dbs.elki.index.preprocessed.snn	Indexes providing nearest neighbor sets
de.lmu.ifi.dbs.elki.index.preprocessed.subspaceproj	Index using a preprocessed local subspaces.
de.lmu.ifi.dbs.elki.index.tree	Tree-based index structures
de.lmu.ifi.dbs.elki.index.tree.metrical	Tree-based index structures for metrical vector spaces.
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants	M-Tree and variants.
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.mktrees	Metrical index structures based on the concepts of the M-Tree supporting processing of reverse k nearest neighbor queries by using the k-nn distances of the entries.
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.mktrees.mkapp	MkAppTree
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.mktrees.mkcop	MkCoPTree
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.mktrees.mkmax	MkMaxTree
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.mktrees.mktab	MkTabTree
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.mtree	MTree
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.query	Classes for performing queries (knn, range, ...) on metrical trees.
de.lmu.ifi.dbs.elki.index.tree.metrical.mtreevariants.split	Splitting strategies of nodes in an M-Tree (and variants).
de.lmu.ifi.dbs.elki.index.tree.query	Classes related to generic tree queries.
de.lmu.ifi.dbs.elki.index.tree.spatial	Tree-based index structures for spatial indexing.
de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants	R*-Tree and variants.
de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants.bulk	Packages for bulk-loading R*-Trees.
de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants.deliclu	DeLiCluTree
de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants.query	Queries on the R-Tree family of indexes: kNN and range queries.
de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants.rstar	RStarTree
de.lmu.ifi.dbs.elki.index.tree.spatial.rstarvariants.util	Utilities for R*-Tree and variants.
de.lmu.ifi.dbs.elki.persistent	Persistent data management.

Datatypes and Distance Functions

de.lmu.ifi.dbs.elki.data	Basic classes for different data types, database object types and label types.
de.lmu.ifi.dbs.elki.data.images	Package for processing image data (e.g. compute color histograms)
de.lmu.ifi.dbs.elki.data.model	Cluster models classes for various algorithms.
de.lmu.ifi.dbs.elki.data.spatial	Spatial data types - interfaces and utilities.
de.lmu.ifi.dbs.elki.data.synthetic	Generators for synthetic data sets
de.lmu.ifi.dbs.elki.data.synthetic.bymodel	Generator using a distribution model specified in an XML configuration file.
de.lmu.ifi.dbs.elki.data.synthetic.bymodel.distribution	Data generators used by the model-based generator.
de.lmu.ifi.dbs.elki.data.type	Data type information, also used for type restrictions.
de.lmu.ifi.dbs.elki.distance	Distance values, distance functions and similarity functions.
de.lmu.ifi.dbs.elki.distance.distancefunction	Distance functions for use within ELKI.
de.lmu.ifi.dbs.elki.distance.distancefunction.adapter	Distance functions deriving distances from e.g. similarity measures
de.lmu.ifi.dbs.elki.distance.distancefunction.colorhistogram	Distance functions using correlations.
de.lmu.ifi.dbs.elki.distance.distancefunction.correlation	Distance functions using correlations.
de.lmu.ifi.dbs.elki.distance.distancefunction.external	Distance functions using external data sources.
de.lmu.ifi.dbs.elki.distance.distancefunction.geo	Geographic (earth) distance functions.
de.lmu.ifi.dbs.elki.distance.distancefunction.subspace	Distance functions based on subspaces.
de.lmu.ifi.dbs.elki.distance.distancefunction.timeseries	Distance functions designed for time series.
de.lmu.ifi.dbs.elki.distance.distancevalue	Distance values, i.e. object storing an actual distance value along with comparison functions and value parsers.
de.lmu.ifi.dbs.elki.distance.similarityfunction	Similarity functions.
de.lmu.ifi.dbs.elki.distance.similarityfunction.kernel	Kernel functions.

Evaluation

de.lmu.ifi.dbs.elki.evaluation	Functionality for the evaluation of algorithms.
de.lmu.ifi.dbs.elki.evaluation.histogram	Functionality for the evaluation of algorithms using histograms.
de.lmu.ifi.dbs.elki.evaluation.index	Simple index evaluation methods
de.lmu.ifi.dbs.elki.evaluation.outlier	Evaluate an outlier score using a misclassification based cost model.
de.lmu.ifi.dbs.elki.evaluation.paircounting	Evaluation of clustering results via pair counting.
de.lmu.ifi.dbs.elki.evaluation.paircounting.generator	Pair generation for pair counting evaluation.
de.lmu.ifi.dbs.elki.evaluation.roc	Evaluation of rankings using ROC AUC (Receiver Operation Characteristics - Area Under Curve)
de.lmu.ifi.dbs.elki.evaluation.similaritymatrix	Render a distance matrix to visualize a clustering-distance-combination.

GUI and Visualization

de.lmu.ifi.dbs.elki.gui	Graphical User Interfaces for ELKI
de.lmu.ifi.dbs.elki.gui.configurator	Configurator components
de.lmu.ifi.dbs.elki.gui.icons	Icons for ELKI GUI.
de.lmu.ifi.dbs.elki.gui.minigui	A very simple UI to build ELKI command lines.
de.lmu.ifi.dbs.elki.gui.multistep	Multi-step GUI for ELKI
de.lmu.ifi.dbs.elki.gui.multistep.panels	Panels for the multi-step GUI
de.lmu.ifi.dbs.elki.gui.util	Utility classes for GUIs (e.g. a class to display a logging panel)
de.lmu.ifi.dbs.elki.visualization	Visualization package of ELKI.
de.lmu.ifi.dbs.elki.visualization.batikutil	Commonly used functionality useful for Apache Batik.
de.lmu.ifi.dbs.elki.visualization.colors	Color scheme handling for ELKI.
de.lmu.ifi.dbs.elki.visualization.css	Managing CSS styles / classes.
de.lmu.ifi.dbs.elki.visualization.gui	Package to provide a visualization GUI.
de.lmu.ifi.dbs.elki.visualization.gui.detail	Classes for managing a detail view.
de.lmu.ifi.dbs.elki.visualization.gui.overview	Classes for managing the overview plot.
de.lmu.ifi.dbs.elki.visualization.opticsplot	Code for drawing OPTICS plots
de.lmu.ifi.dbs.elki.visualization.projections	Visualization projections
de.lmu.ifi.dbs.elki.visualization.projector	Projectors are responsible for finding appropriate projections for data relations.
de.lmu.ifi.dbs.elki.visualization.savedialog	Save dialog for SVG plots.
de.lmu.ifi.dbs.elki.visualization.scales	Scales handling for plotting.
de.lmu.ifi.dbs.elki.visualization.style	Style management for ELKI visualizations.
de.lmu.ifi.dbs.elki.visualization.style.lines	Generate line styles for plotting in CSS
de.lmu.ifi.dbs.elki.visualization.style.marker	Draw plot markers
de.lmu.ifi.dbs.elki.visualization.svg	Base SVG functionality (generation, markers, thumbnails, export, ...).
de.lmu.ifi.dbs.elki.visualization.visualizers	Visualizers for various results
de.lmu.ifi.dbs.elki.visualization.visualizers.events	Events occuring in visualization contexts
de.lmu.ifi.dbs.elki.visualization.visualizers.optics	Visualizers that do work on OPTICS plots
de.lmu.ifi.dbs.elki.visualization.visualizers.parallel	Visualizers based on parallel coordinates.
de.lmu.ifi.dbs.elki.visualization.visualizers.thumbs	Thumbnail "Visualizers" (that take care of refreshing thumbnails)
de.lmu.ifi.dbs.elki.visualization.visualizers.vis1d	Visualizers based on 1D projections.
de.lmu.ifi.dbs.elki.visualization.visualizers.vis2d	Visualizers based on 2D projections.
de.lmu.ifi.dbs.elki.visualization.visualizers.visunproj	Visualizers that do not use a particular projection.

Utilities and Miscellaneous

de.lmu.ifi.dbs.elki	ELKI framework "Environment for Developing KDD-Applications Supported by Index-Structures" KDDTask is the main class of the ELKI-Framework for command-line interaction.
de.lmu.ifi.dbs.elki.application	Base classes for stand alone applications.
de.lmu.ifi.dbs.elki.application.cache	Utility applications for the persistence layer such as distance cache builders.
de.lmu.ifi.dbs.elki.application.internal	Internal utilities for development.
de.lmu.ifi.dbs.elki.application.jsmap	JavaScript based map client - server architecture.
de.lmu.ifi.dbs.elki.application.visualization	Visualization applications in ELKI.
de.lmu.ifi.dbs.elki.logging	Logging facility for controlling logging behavior of the complete framework.
de.lmu.ifi.dbs.elki.logging.progress	Progress status objects (for UI)
de.lmu.ifi.dbs.elki.math	Mathematical operations and utilities used throughout the framework.
de.lmu.ifi.dbs.elki.math.linearalgebra	Linear Algebra package provides classes and computational methods for operations on matrices.
de.lmu.ifi.dbs.elki.math.linearalgebra.fitting	Function to numerically fit a function (such as a Gaussian distribution to given data.
de.lmu.ifi.dbs.elki.math.linearalgebra.pca	Principal Component Analysis (PCA) and Eigenvector processing.
de.lmu.ifi.dbs.elki.math.linearalgebra.pca.weightfunctions	Weight functions used in weighted PCA via WeightedCovarianceMatrixBuilder
de.lmu.ifi.dbs.elki.math.spacefillingcurves	Space filling curves.
de.lmu.ifi.dbs.elki.math.statistics	Statistical tests and methods.
de.lmu.ifi.dbs.elki.properties	Property handling and main ELKI properties file.
de.lmu.ifi.dbs.elki.result	Result types, representation and handling
de.lmu.ifi.dbs.elki.result.optics	Result classes for OPTICS.
de.lmu.ifi.dbs.elki.result.outlier	Outlier result classes
de.lmu.ifi.dbs.elki.result.textwriter	Text serialization (CSV, Gnuplot, Console, ...)
de.lmu.ifi.dbs.elki.result.textwriter.naming	Naming schemes for clusters (for output when an algorithm doesn't generate cluster names).
de.lmu.ifi.dbs.elki.result.textwriter.writers	Serialization handlers for individual data types.
de.lmu.ifi.dbs.elki.utilities	Utility and helper classes - commonly used data structures, output formatting, exceptions, ...
de.lmu.ifi.dbs.elki.utilities.datastructures	Basic memory structures such as heaps and object hierarchies.
de.lmu.ifi.dbs.elki.utilities.datastructures.heap	Heap structures and variations such as bounded priority heaps.
de.lmu.ifi.dbs.elki.utilities.datastructures.hierarchy	Delegate implementation of a hierarchy.
de.lmu.ifi.dbs.elki.utilities.designpattern	Interfaces and implementations related to common "design patterns".
de.lmu.ifi.dbs.elki.utilities.documentation	Documentation utilities: Annotations for Title, Description, Reference
de.lmu.ifi.dbs.elki.utilities.exceptions	Exception classes and common exception messages.
de.lmu.ifi.dbs.elki.utilities.iterator	Various Iterator decorators and adapters.
de.lmu.ifi.dbs.elki.utilities.optionhandling	Parameter handling and option descriptions.
de.lmu.ifi.dbs.elki.utilities.optionhandling.constraints	Constraints allow to restrict possible values for parameters.
de.lmu.ifi.dbs.elki.utilities.optionhandling.parameterization	Configuration managers See the de.lmu.ifi.dbs.elki.utilities.optionhandling package for documentation!
de.lmu.ifi.dbs.elki.utilities.optionhandling.parameters	Classes for various typed parameters.
de.lmu.ifi.dbs.elki.utilities.pairs	Pairs and triples utility classes.
de.lmu.ifi.dbs.elki.utilities.referencepoints	Package containing strategies to obtain reference points Shared code for various algorithms that use reference points.
de.lmu.ifi.dbs.elki.utilities.scaling	Scaling functions: linear, logarithmic, gamma, clipping, ...
de.lmu.ifi.dbs.elki.utilities.scaling.outlier	Scaling of Outlier scores, that require a statistical analysis of the occurring values
de.lmu.ifi.dbs.elki.utilities.xml	XML and XHTML utilities.
de.lmu.ifi.dbs.elki.workflow	Work flow packages, e.g. following the usual KDD model, closely related to CRISP-DM

ELKI: Environment for DeveLoping KDD-Applications Supported by Index-Structures.

ELKI is a generic framework for a broad range of KDD-applications and their development. For background, contact-information, and contributors see http://www.dbs.ifi.lmu.de/research/KDD/ELKI/.

This is the documentation for version 0.4, published as:
Elke Achtert, Ahmed Hettab, Hans-Peter Kriegel, Erich Schubert, Arthur Zimek:
Spatial Outlier Detection: Data, Algorithms, Visualizations.
12th International Symposium on Spatial and Temporal Databases (SSTD), Minneapolis, MN, 2011.

Getting started

The ELKI wiki has additional documentation and will continuously be updated. A Tutorial exported from the Wiki is included with this documentation and a good place to start.

Invocation

To use the KDD-Framework we recommend an executable .jar-file: elki.jar. Since release 0.3 it will by default invoke a minimalistic GUI called MiniGUI when you call java -jar elki.jar. For command line use (for example for batch processing and scripted operation), you can get a description of usage by calling java -cp elki.jar de.lmu.ifi.dbs.elki.application.KDDCLIApplication -h.

The MiniGUI can also serve as a utility for building command lines, as it will print the full command line to the log window.

For more information on using files and available formats as data input see de.lmu.ifi.dbs.elki.datasource.parser. ELKI 0.4 uses a whitespace separated vector format by default, but there also is a parser for ARFF files included that can read most ARFF files (mixing sparse and dense vectors is currently not allowed).

An extensive list of parameters can be browsed sorted by class or sorted by option ID.

Some examples of completely parameterized calls for different algorithms are described at example calls.

A list of related publications, giving details on many implemented algorithms, can be found in the class article references list.

Workflow - Where Do Which Objects Go?

The database connection manages reading of input files or databases and provides a Database-Object - including index structures - as a virtual database to the KDDTask. The KDDTask applies a specified algorithm on this database and collects the result from the algorithm. Finally, KDDTask hands on the obtained result to a ResultHandler. The default-handler is ResultWriter, writing the result to STDOUT or, if specified, into a file.

Database and indexing layer

The database and indexing layer is a key component of ELKI. This is not just a storage for double[], as with many other frameworks. It can store various types of objects, and the integrated index structures provide access to fast distance, similarity, kNN, RkNN and range query methods for a variety of distance functions.

The standard flow for initializing a database is as depicted here:

The standard stream-based data sources such as FileBasedDatabaseConnection will open the stream, feed the contents through a Parser to obtain an initial MultipleObjectsBundle. This is a temporary container for the data, which can then be modified by arbitrary ObjectFilters.
In the end, the MultipleObjectsBundle is bulk-inserted into a Database, which will then invoke its IndexFactorys to add Index instances to the appropriate relations.

When a request for a distance, similarity, kNN, RkNN or range query is received by the database, it queries all indexes if they have support for this query. If so, an optimized query is returned, otherwise a linear scan query can be returned unless DatabaseQuery.HINT_OPTIMIZED_ONLY was given.

For this optimization to work, you should be using the proper APIs of the Database interface or QueryUtil helper where possible, instead of initializing low level classes such as an explicit linear scan query.

For efficiency, try to instantiate the query only once per algorithm run, and avoid running the optimization step for every object.

How to make use of this framework

Extension

To provide new applications one is simply to implement the specified interfaces. There are interfaces for a broad range of targets of development. Compare the tree of interfaces to get an overview concerning the provided interfaces.

A good place to get started is to have a look at some of the existing algorithms, and see how they are implemented. For example the DummyAlgorithm while it does not produce any result, will teach you how to perform k-nearest-neighbor queries properly. It does however have a hard dependency on the Euclidean distance and the datatypes supported by it. In order to support arbitrary distance functions, extend the class AbstractDistanceBasedAlgorithm instead. This is another simple example, this time for obtaining a class parameter.

Visit the ELKI Wiki, which has a growing amount of documentation. You are also welcome to contribute, of course!

Parameterization API

ELKI is designed for command-line, GUI and Java operation. For command-line and GUI, an extensive help functionality is provided along with input assistance. Therefore, you should also support the parameterizable API. The requirements are quite different from regular Java constructors, and cannot be expressed in terms of a Java API.

For useful error reporting and input assistance in the GUI we need to have more extensive typing than Java uses (for example we might need numerical constraints) and we also want to be able to report more than one error at a time. In ELKI 0.4, much of the parameterization was refactored to static helper classes usually found as a public static class Parameterizer and subclasses of AbstractParameterizer.

Keep the complexity of Parameterizer classes and constructors invoked by these classes low, since these may be heavily used during the parameterization step. Postpone any extensive initialization to the main algorithm invocation step!