Download JAR files tagged by scale with all dependencies

A Functional Workflow Language. Cuneiform is a workflow specification language which makes it easy to integrate heterogeneous tools and libraries and exploit data parallelism. Users do not have to create heavy-weight wrappers for establised tools or to reimplement them. Instead, they apply their existing software to partitioned data. Using the Hi-WAY application master Cuneiform can be executed on Hadoop YARN which makes it suitable for large scale data analysis. Cuneiform comes in the form of a functional programming language with a Foreign Function Interface (FFI) that lets users create functions in any suitable scripting language and apply these functions in a uniform way. Data paralelism is expressed by applying map, cross-product, dot-product, or combinations of the aforementioned algorithmic skeletons to collections of black-box data.

Extension of BestFirst. Takes a restricted number of k attributes into account. Fixed-set selects a fixed number k of attributes, whereas k is increased in each step when fixed-width is selected. The search uses either the initial ordering to select the top k attributes, or performs a ranking (with the same evalutator the search uses later on). The search direction can be forward, or floating forward selection (with opitional backward search steps). For more information see: Martin Guetlein (2006). Large Scale Attribute Selection Using Wrappers. Freiburg, Germany.

A tool to help visualise API differences between two different versions of a project. Jardiff takes two jar files and outputs all the public API changes as xml, html or plain text. Currently this is command line driven, though we are working on an ant plugin. It requires java 1.4 to run. <h2>Background</h2> At OSjava, we were using a shell script to work out API differences between our release versions. This unfortunately doesn't scale very well to larger projects, the main reason for jardiff was to do the same in a much shorter time (From memory it was taking minutes to get the API diff from two fairly small jars). In it's current state, jardiff can generate the api differences between rt.jar for java 1.4 and java 1.5 in under 20 seconds, so it should scale to even the largest of projects.

OpenIMAJ (Open Intelligent Multimedia in Java) is a collection of libraries and tools for multimedia analysis written in the Java programming language. OpenIMAJ intends to be the first truly complete multimedia analysis library and contains modules for analysing images, videos, text, audio and even webpages. The OpenIMAJ image and video analysis and feature extraction modules contain methods for processing visual content and extracting state-of-the-art features, including SIFT. The OpenIMAJ clustering and nearest-neighbour libraries contain efficient, multi-threaded implementations of clustering algorithms including Hierarchical K-Means and Approximate K-Means. The clustering library makes it possible to easily create visual-bag-of-words representations for images and video with very large vocabularies. The text-analysis modules contain implementations of a statistical language classifier and low-level processing pipeline. A number of modules deal with content creation, including interactive slideshows and animations. The hardware integration modules allow cross-platform integration with devices including webcams, the Microsoft Kinect, and even devices such as GPS's. OpenIMAJ also incorporates a number of tools to enable extremely-large-scale multimedia analysis using a distributed computing approach based on Apache Hadoop.

The OODT grid services (product and profile services) use CORBA or RMI as their underlying network transport. However, limitations of CORBA and RMI make them inappropriate for large-scale deployments. For one, both are procedural mechanisms, providing a remote interface that resembles a method call. This makes streaming of data from a service impossible, because there are limitations to the sizes of data structures that can be passed over a remote method call. Instead, repeated calls must be made to retrieve each block of a product, making transfer speeds horribly slow compared to HTTP or FTP. (Block-based retrieval of profiles was never implemented, resulting in out of memory conditions for large profile results, which is another problem.) Second, both CORBA and RMI rely on a central name registry. The registry makes an object independent of its network location, enabling a client to call it by name (looking up its last known location in the registry). However, this requires that server objects be able to make outbound network calls to the registry (through any outbound firewall), and that the registry accept those registrations (through any inbound firewall). This required administrative action at institutions hosting server objects and at the institution hosting the registry. Often, these firewall exceptions would change without notice as system adminstrators changed at each location (apparently firewall exceptions are poorly documented everywhere). Further, in the two major deployments of OODT (PDS and EDRN), server objects have almost never moved, nullifying any benefit of the registry. This project, OODT Web Grid Services, avoids the prolems of CORBA and RMI by using HTTP as the transport mechanism for products and profiles. Further, it provides a password-protected mechanism to add new sets of product and profile query handlers, enabling seamless activation of additional capabilities.