Updates 2024/07/17

[Peter Reutemann]

Hi everyone

This time round, there have been a bunch of important bug-fixes and additions. The Weka Investigator can now perform repeated cross-validation, calculating statistics across the runs (the Experimenter does this per fold, the Investigator per full CV) and also a new tab for executing an experiment run on the current dataset and a chosen classifier.

Last, but not least, ADAMS turned 15 today (first commit in 2009). :-)

Fixes

• adams-pdf: PDF fonts now properly parse the attributes in the font string (Arrays.binarySearch on unsorted array returned invalid index for Italic).

• adams-weka and adams-weka-lts:

  • The Multi-Experimenter now sorts the results according to dataset/classifier order (dataset first).

  • Leave-one-out cross-validation (LOOCV) now works correctly in the Multi-Experimenter.

  • The Rename... action for renaming a column of an InstancesTable now uses the correct column index.

  • Changed the R^2 calculation to https://en.wikipedia.org/wiki/Coefficient_of_determination#Definitions

Changes

• adams-core:

  • spreadsheet formulas can now use median as well

  • upgraded commons-io to 2.16.1

• adams-imaging:

  • upgraded opex4j to 0.0.3

  • the ObjectCentersFromReport plugin in the Preview browser now allows changing the report reader

  • object finders now have a quiet flag to suppress warnings about -1 indices

• adams-spreadsheet:

  • The SpreadSheetAggregate transformer now offers the 25th and 75th quartile as well.

• adams-weka and adams-weka-lts:

  • upgraded common-csv-weka-package to 2024.7.5

Additions

• adams-imaging:

  • The image object filter ObjectsInPolygon finds annotations that fall within the specified polygon.

  • The CountObjectsInPolygon transformer outputs the number of annotations that fall within the specified polygon.

• adams-weka and adams-weka-lts:

  • Added support for repeated cross-validation to the Weka Investigator.

  • Added an Experiment tab to the Weka Investigator for evaluating single classifier on selected dataset.

Cheers, Peter

--
Peter Reutemann
Dept. of Computer Science
University of Waikato, Hamilton, NZ
Mobile +64 22 190 2375
https://profiles.waikato.ac.nz/peter.reutemann
http://www.data-mining.co.nz/

[frac...@waikato.ac.nz]

> This time round, there have been a bunch of important bug-fixes and
> additions. The Weka Investigator can now perform repeated
> cross-validation, calculating statistics across the runs (the
> Experimenter does this per fold, the Investigator per full CV) and also
> a new tab for executing an experiment run on the current dataset and a
> chosen classifier.

I created short videos that demonstrate the new features:
- repeated CV: https://youtu.be/ZlbynLCq8mc?si=vTPcM_7Ehtot2Kk6
- experiment tab: https://youtu.be/bVid9WrpE0s?si=QjO1FfrCuwnYhZ8u