TAR2 is a treatment learner, or more specificly, a data mining/summarization tool. Both treatment learners and other machine learners are rule discovery paradigms. However, classical machine learners like C4.5 aim at discovering classification rules: i.e. given a classified training set, they output rules that are predictive of the class attribute. TAR2 differs from those learners in that:

• TAR2 assumes the classes are ordered by their scores(some domain-specific measure).
• Highly scored classes are preferable to lower scored classes.
• Further, one class is more desirable than all others, which is called the best class.
• Rather than finding the classification rules, TAR2 finds rules that predict both increased frequency of the best class and decreased frequency of the worst class.

TAR2 inputs classified data logs and output treatments. A treatment is one or a conjunction of attribute values. It is a constraint on future controllable inputs of the system. In summary, treatment learners give us controllers rather than classifiers. To understand the distinction, consider the case of someone reading a map. Classifiers say "you are here" on the map while controllers say "go this way".

Simply unzip tar2.zip and you get the following:

• Source code of TAR2 and a X-way cross validation facility.
• All DOS executables to run TAR2 and X-way cross validation experiment (we are told, but can't confirm, that rebuilding this for UNIX is just a matter of cd-ing to the source directory and just typing "make").
• Sample datasets and their output files.
• Documents including instructions and several associated research papers.

• Go into tar2\bin and type: tar2 filestem tar2 filestem > filestem.out.
  Example: tar2 c:\tar2\samples\iris\iris > iris.out
• Also, you can make a batch file with that one line command and run TAR2 using your mouse.

• Make sure each of the three files: filestem.data, filestem.names, filestem.cfg is of the correct format.
• Filestem should not be "XDF" when running tar2.
• First run: set all the parameters to default,TAR2 prints out the deltaf distribution of the dataset.
• After first run: set promising to a non-zero value according to the deltaf distribution (generally, set promising to a larger deltaf value, it can be a decimal) and run TAR2 once more to get treatments.
• Increase nchanges to see if the result is improving. Generally, nchanges is less than 4.
• Use skew to control the size of result set: some subsets that satisfy certain treatments may be too small to be convincing. Set skew = N to report only subsets that contain at least 1/N cases with best class of the original set.

TARZAN (a.k.a. TAR1)

Tim Menzies with help from Erik Sinsel

TAR2 (alpha):

Tim Menzies.
Awful Prolog prototype. Barely usable.

TAR2, TAR3

Ying Hu (with remote and contradictory advice from Tim Menzies).
Runs fast. Simple to use