Download tests JAR files with dependency

Classification by voting feature intervals. Intervals are constucted around each class for each attribute (basically discretization). Class counts are recorded for each interval on each attribute. Classification is by voting. For more info see: G. Demiroz, A. Guvenir: Classification by voting feature intervals. In: 9th European Conference on Machine Learning, 85-92, 1997.

Simple learning schemes for educational purposes (Prism, Id3, IB1 and NaiveBayesSimple).

Class implementing minimal cost-complexity pruning. Note when dealing with missing values, use "fractional instances" method instead of surrogate split method. For more information, see: Leo Breiman, Jerome H. Friedman, Richard A. Olshen, Charles J. Stone (1984). Classification and Regression Trees. Wadsworth International Group, Belmont, California.

Cluster data using the sequential information bottleneck algorithm. Note: only hard clustering scheme is supported. sIB assign for each instance the cluster that have the minimum cost/distance to the instance. The trade-off beta is set to infinite so 1/beta is zero. For more information, see: Noam Slonim, Nir Friedman, Naftali Tishby: Unsupervised document classification using sequential information maximization. In: Proceedings of the 25th International ACM SIGIR Conference on Research and Development in Information Retrieval, 129-136, 2002.

Wrapper classifiers for Jython and Groovy code. Even though the classifier is serializable, the trained classifier cannot be stored persistently. I.e., one cannot store a model file and re-load it at a later point in time again to make predictions.

An implementation of a RIpple-DOwn Rule learner. It generates a default rule first and then the exceptions for the default rule with the least (weighted) error rate. Then it generates the "best" exceptions for each exception and iterates until pure. Thus it performs a tree-like expansion of exceptions.The exceptions are a set of rules that predict classes other than the default. IREP is used to generate the exceptions. For more information about Ripple-Down Rules, see: Brian R. Gaines, Paul Compton (1995). Induction of Ripple-Down Rules Applied to Modeling Large Databases. J. Intell. Inf. Syst. 5(3):211-228.

Class for boosting a 2-class classifier using the Real Adaboost method. For more information, see J. Friedman, T. Hastie, R. Tibshirani (2000). Additive Logistic Regression: a Statistical View of Boosting. Annals of Statistics. 95(2):337-407.

Classifier for incremental learning of large datasets by way of racing logit-boosted committees. For more information see: Eibe Frank, Geoffrey Holmes, Richard Kirkby, Mark Hall: Racing committees for large datasets. In: Proceedings of the 5th International Conferenceon Discovery Science, 153-164, 2002.

Races the cross validation error of competing attribute subsets. Use in conjuction with a ClassifierSubsetEval. RaceSearch has four modes: forward selection races all single attribute additions to a base set (initially no attributes), selects the winner to become the new base set and then iterates until there is no improvement over the base set. Backward elimination is similar but the initial base set has all attributes included and races all single attribute deletions. Schemata search is a bit different. Each iteration a series of races are run in parallel. Each race in a set determines whether a particular attribute should be included or not---ie the race is between the attribute being "in" or "out". The other attributes for this race are included or excluded randomly at each point in the evaluation. As soon as one race has a clear winner (ie it has been decided whether a particular attribute should be inor not) then the next set of races begins, using the result of the winning race from the previous iteration as new base set. Rank race first ranks the attributes using an attribute evaluator and then races the ranking. The race includes no attributes, the top ranked attribute, the top two attributes, the top three attributes, etc. It is also possible to generate a raked list of attributes through the forward racing process. If generateRanking is set to true then a complete forward race will be run---that is, racing continues until all attributes have been selected. The order that they are added in determines a complete ranking of all the attributes. Racing uses paired and unpaired t-tests on cross-validation errors of competing subsets. When there is a significant difference between the means of the errors of two competing subsets then the poorer of the two can be eliminated from the race. Similarly, if there is no significant difference between the mean errors of two competing subsets and they are within some threshold of each other, then one can be eliminated from the race.

Class for building pace regression linear models and using them for prediction. Under regularity conditions, pace regression is provably optimal when the number of coefficients tends to infinity. It consists of a group of estimators that are either overall optimal or optimal under certain conditions. The current work of the pace regression theory, and therefore also this implementation, do not handle: - missing values - non-binary nominal attributes - the case that n - k is small where n is the number of instances and k is the number of coefficients (the threshold used in this implmentation is 20) For more information see: Wang, Y (2000). A new approach to fitting linear models in high dimensional spaces. Hamilton, New Zealand. Wang, Y., Witten, I. H.: Modeling for optimal probability prediction. In: Proceedings of the Nineteenth International Conference in Machine Learning, Sydney, Australia, 650-657, 2002.