Download JAR files tagged by gradient with all dependencies

This module defines the interfaces for Comparators in the TCS Alignment Toolbox. A Comparator has the purpose of defining the dissimilarity between elements in the input sequences of an Alignment. More specific information on Comparators can be found in the 'Comparator' interface. You can find a lot of helpful standard implementations of Comparators in the comparators-lib module. In the TCS Alignment Toolbox we require the output values of Comparators to lie in the range [0,1]. Many natural dissimilarities on value sets do not meet this criterion, such that additional normalization has to be applied. To that end this package also contains a Normalizer interface for functions that map real values from the range [0, infinity) to the range [0,1]. This package also provides a few convenience implementations of the Comparator interface to make the implementation of custom Comparators simpler, namely: SkipExtendedComparator, ParameterLessSkipExtendedComparator, ComparisonBasedSkipExtendedComparator, and ParameterLessComparisonBasedSkipExtendedComparator. Finally the TCS Alignment Toolbox also provides the means to learn parameters of Comparators. To enable that Comparators must implement the DerivableComparator interface to properly define the parameters that can be learned and the gradient of the dissimilarity with respect to these parameters. Gradients are stored using the Gradient interface as well as some convenience implementations of said interface, namely EmptyGradient, SingletonGradient, ArrayGradient and ListGradient.

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.