What's new or improved in Weka 3.7.4

Core Weka

  • SimpleKMeans now has an option to initialize clusters using the k-means++ method.
  • GUI and command-line package manager clients now report download progress.
  • TextDirectoryLoader now allows a character set to be specified for reading text files.
  • CSVLoader now supports reading of date fields.
  • Vote meta classifier now has an option to load serialized models and include them in the ensemble.
  • Major UI and functionality overhaul for the Knowledge Flow (multiple flows open in separate tabs, new editing functionality, parallel and sequential execution of separate paths, built in templates, "notes" on the canvas, file associations for flow files, plugin "perspectives" etc.).
  • Built-in scatter plot matrix, attribute summarizer and SQL viewer perspectives for the Knowledge Flow.

In Packages

  • scatterPlot3D now includes a perspective for the Knowledge Flow
  • new timeseriesForecasting package for performing closed-loop forecasting with any Weka regression scheme as base learner. Includes evaluation routines, JFreeChart-based visualization, an Explorer plugin, a Knowledge Flow perspective, and a component for using forecasting models in a Knowledge Flow process.
  • Improved performance in the WekaODF open document reader/writer package.
  • TPP - projection pursuit visualization Explorer plugin contributed by Joe Faith from Northumbria University.