Update 2016/05/20

A major change in versioning of the artifacts has occurred: from now on, artifacts will have YY.MM.x as version, with YY being the year, MM the month and x the release in that month, starting with 0. This makes it a bit easier tracking down from what time period a snapshot is.

Fixes

  • adams-weka: The WekaExtractArray transformer now uses the correct maximum for columns/rows in case of spreadsheets (was accidentally swapped).
  • Default timezone no longer results in using GMT but the system-defined one. Java's timezone uses GMT if a timezone string cannot be parsed.

Changes

  • adams-core:
    • The Injector control actor can replace tokens passing through now, not just adding them before or after. Furthermore, it can also inject data from storage now.
    • The TrainableBatchFilter interface is used by transformers derived from AbstractFilter to train them if necessary with the first batch of data. Such filters can be reset with the monitor variable of the transformer.
  • adams-weka: upgraded multisearch-weka-package to 2016.5.15, which fixes a multithreading bug
  • Moved the following modules from base to addons:
    • gnuplot
    • latex
    • maps
    • moa
    • osm
    • r
    • random
    • twitter

Additions

  • adams-spreadsheet: It is now possible to extract columns or rows from a spreadsheet by using the SpreadSheetExtractArray transformer.
  • Added new addons module adams-dl4j, which makes the deeplearning4j library available within ADAMS.