Download JAR files tagged by ridge with all dependencies

Performs a grid search of parameter pairs for the a classifier (Y-axis, default is LinearRegression with the "Ridge" parameter) and the PLSFilter (X-axis, "# of Components") and chooses the best pair found for the actual predicting. The initial grid is worked on with 2-fold CV to determine the values of the parameter pairs for the selected type of evaluation (e.g., accuracy). The best point in the grid is then taken and a 10-fold CV is performed with the adjacent parameter pairs. If a better pair is found, then this will act as new center and another 10-fold CV will be performed (kind of hill-climbing). This process is repeated until no better pair is found or the best pair is on the border of the grid. In case the best pair is on the border, one can let GridSearch automatically extend the grid and continue the search. Check out the properties 'gridIsExtendable' (option '-extend-grid') and 'maxGridExtensions' (option '-max-grid-extensions <num>'). GridSearch can handle doubles, integers (values are just cast to int) and booleans (0 is false, otherwise true). float, char and long are supported as well. The best filter/classifier setup can be accessed after the buildClassifier call via the getBestFilter/getBestClassifier methods. Note on the implementation: after the data has been passed through the filter, a default NumericCleaner filter is applied to the data in order to avoid numbers that are getting too small and might produce NaNs in other schemes.

This package currently contains classes for training multilayer perceptrons with one hidden layer, where the number of hidden units is user specified. MLPClassifier can be used for classification problems and MLPRegressor is the corresponding class for numeric prediction tasks. The former has as many output units as there are classes, the latter only one output unit. Both minimise a penalised squared error with a quadratic penalty on the (non-bias) weights, i.e., they implement "weight decay", where this penalised error is averaged over all training instances. The size of the penalty can be determined by the user by modifying the "ridge" parameter to control overfitting. The sum of squared weights is multiplied by this parameter before added to the squared error. Both classes use BFGS optimisation by default to find parameters that correspond to a local minimum of the error function. but optionally conjugated gradient descent is available, which can be faster for problems with many parameters. Logistic functions are used as the activation functions for all units apart from the output unit in MLPRegressor, which employs the identity function. Input attributes are standardised to zero mean and unit variance. MLPRegressor also rescales the target attribute (i.e., "class") using standardisation. All network parameters are initialised with small normally distributed random values.