Torch Reference Manual

http://www.torch.ch

Table of Contents

General stuff

• List
• Text outputs functions.

Namespaces

• Givens matrix operations routines.
• Householder transformation routines.
• Routines for determining Hessenberg factorisations.
• Routines for symmetric eigenvalue problems.
• Collection of matrix factorisation operation functions.
• Collection of matrix operation functions.
• Collection of permutations operation functions.
• Some simple functions for log operations.
• Some simple functions for string operations.

Classes

• Allocator Class do easily allocate/deallocate memory in Torch.
• Bagging This class represents a Trainer that implements the well-known Bagging algorithm (Breiman, 1996).
• BayesClassifier A multi class bayes classifier -- maximizes the likelihood of each class separately using a trainer for distribution.
• BayesClassifierMachine BayesClassifierMachine is the machine used by the BayesClassifier trainer to perform a Bayes Classification using different distributions.
• BoolCmdOption This class defines a bool command-line option.
• Boosting Boosting implementation.
• BoostingMeasurer Compute the classification weighted error (in %) for BoostingMachine of the inputs with respect to the targets of data.
• ClassFormat Used to define a class code.
• ClassFormatDataSet Given a DataSet, convert (on-the-fly) targets using a conversion table.
• ClassMeasurer Compute the classification error (in %) of the inputs with respect to the targets of data.
• ClassNLLCriterion This criterion can be used to train *in classification* a GradientMachine object using the StochasticGradient trainer.
• ClassNLLMeasurer This class measures the negative log likelihood.
• CmdLine This class provides a useful interface for the user, to easily read some arguments/options from the command-line.
• CmdOption This class defines an option for the command line.
• ConnectedMachine Easy connections between several GradientMachine.
• Criterion Criterion class for StochasticGradient.
• DataSet Provides an interface to manipulate all kind of data.
• DiagonalGMM This class can be used to model Diagonal Gaussian Mixture Models.
• DiskDataSet Provides an interface to manipulate all kind of data which are kept on disk, and not fully loaded in memory.
• DiskHTKDataSet Provides an interface to manipulate HTK data which are kept on disk, and not fully loaded in memory.
• DiskMatDataSet Matrix DataSet On Disk.
• DiskXFile A file on the disk.
• Distribution This class is designed to handle generative distribution models such as Gaussian Mixture Models and Hidden Markov Models.
• DotKernel DotProduct Kernel.
• EMTrainer This class is used to train any distribution using the EM algorithm.
• EditDistance This class can be used to compute the "edit distance" between two sequences.
• EditDistanceMeasurer This class can be used to measure and print an EditDistance object.
• ExampleFrameSelectorDataSet This dataset is empty at the begining.
• Exp Exponentiel layer for GradientMachine.
• FileListCmdOption This class take a file name in the command line, and reads a list of files contained in this file.
• FrameSelectorDataSet A dataset used to select some frames of another dataset.
• GaussianKernel Gaussian Kernel
• GradientMachine Gradient machine: machine which can be trained with a gradient descent.
• Grammar This class contains the grammar of accepted sentences for a speech recognition experiment such as the one using SimpleDecoderSpeechHMM A grammar is a transition table where each node is a word.
• HMM This class implements a Hidden Markov Model distribution.
• HTKDataSet This dataset can deal with the HTK format for features and targets.
• IOAscii Handles the standard Ascii sequence format in Torch.
• IOBin Handles the standard binary sequence format in Torch.
• IOBufferize This IO bufferizes the asked sequence of a given IO when calling getSequence().
• IOHTK Handles the standard binary sequence format in HTK.
• IOHTKTarget Handles the standard Ascii HTK targets/labels format in Torch.
• IOMulti This IO takes several IOSequence, and will act as if you had concatened all these IOSequence when calling getMatrix().
• IOSequence Class which provides an ensemble of sequences, which have the same frame size, but could have different number of frames.
• IOSequenceArray Load and save in an efficiently manner an array of sequences.
• IOSub IOSequence which does a selection of adjacent columns on another IOSequence, when calling getMatrix().
• InputsSelect Machine which select a block of adjacent inputs, and put them in the outputs.
• IntCmdOption This class defines a integer command-line option.
• KFold Provides an interface to sample data, for use by methods such as cross-validation
• KMeans This class can be used to do a "kmeans" on a given set of data.
• KNN This machine implements the K-nearest-neighbors (KNN) algorithm.
• Kernel General Kernel class.
• LexiconInfo This class stores information about how phonemes are assembled into pronunciations.
• Linear Linear layer for GradientMachine.
• LogMixer LogMixer useful for experts mixtures.
• LogRBF LogRBF layer for GradientMachine.
• LogSigmoid Log-Sigmoid layer for GradientMachine.
• LogSoftMax Log-SoftMax layer for GradientMachine.
• MAPDiagonalGMM This class is a special case of a DiagonalGMM that implements the MAP algorithm instead of the EM algorithm.
• MAPHMM This class is a special case of a HMM that implements the MAP algorithm for HMM transitions probabilities.
• MLP A Multi-Layer Perceptron.
• MSECriterion Mean Squared Error criterion.
• MSEMeasurer Mean Squared Error measurer.
• Machine Object which can compute some outputs, given some inputs.
• Mat Matrix object.
• MatDataSet Matrix DataSet.
• MeanVarNorm In the constructor, it computes the mean and the standard deviation over all the frames in the given DataSet (by default, only for inputs).
• Measurer Used to measure what you want during training/testing.
• MemoryDataSet DataSet where data is fully loaded in memory.
• MemoryXFile A file in the memory.
• Mixer Mixer useful for experts mixtures.
• MultiClassFormat Define the multi class encoding format.
• MultiCriterion MultiCriterion can be used to handle multiple criterions.
• Multinomial This class can be used to model Multinomial Distributions.
• NLLCriterion This criterion can be used to train Distribution object using the GMTrainer trainer.
• NLLMeasurer This class measures the negative log likelihood.
• NPTrainer Trainer for Non Parametric Machines.
• NullXFile NullXFile.
• Object Almost all classes in Torch should be a sub-class of this class.
• OneHotClassFormat Define the one hot class encoding format.
• OutputMeasurer Compute the outputification error (in %) of the inputs with respect to the targets of data.
• Parameters Parameters definition.
• ParzenDistribution This class can be used to model a Parzen density estimator with a Gaussian kernel:
• ParzenMachine This machine implements the Parzen Window estimator.
• Perm Permutation object.
• PhoneInfo This class contains the names of the phonemes that make up the words in the lexicon.
• PipeXFile A file in a pipe.
• PolynomialKernel Polynomial Kernel .
• PreProcessing This class is able to do pre-processing on examples in a DataSet.
• QCCache "Cache" used by the Quadratic Constrained Trainer (QCTrainer).
• QCMachine "Quadratic Constrained Machine".
• QCTrainer Train a QCMachine.
• Random Random class which contains several static random methods.
• RealCmdOption This class defines a real command-line option.
• SVM Support Vector Machine.
• SVMCache QCCache implementation for SVMs.
• SVMCacheClassification Cache for SVM classification.
• SVMCacheRegression Cache for SVM regression.
• SVMClassification SVM in classification.
• SVMRegression SVM in regression.
• SaturationMeasurer Measure the saturation of a GradientMachine.
• Sequence Sequence definition.
• Sigmoid Sigmoid layer for GradientMachine.
• SigmoidKernel Sigmoid Kernel
• SimpleDecoderSpeechHMM This class implements a special case of Hidden Markov Models that can be used to do connected word speech recognition for small vocabulary, using embedded training.
• SoftMax SoftMax layer for GradientMachine.
• SoftPlus SoftPlus layer for GradientMachine.
• SpatialConvolution Class for doing convolution over images.
• SpatialSubSampling Class for doing sub-sampling over images.
• SpeechHMM This class implements a special case of Hidden Markov Models that can be used to do connected word speech recognition for small vocabulary, using embedded training.
• Stack This is an implementation of a "stack".
• StochasticGradient Trainer for GradientMachine.
• StringCmdOption This class defines a string command-line option.
• SumMachine This machine simply adds up its input vectors.
• TableLookupDistribution This class outputs one of the observations as the logProbability.
• Tanh Tanh layer for GradientMachine.
• TemporalConvolution Class for doing a convolution over a sequence.
• TemporalMean Given an input sequence, it does the mean over all input frames.
• TemporalSubSampling Class for doing sub-sampling over a sequence.
• TimeMeasurer Measure the time (in seconds) between two measureIteration() calls.
• Timer Timer.
• Trainer Trainer.
• TwoClassFormat Define the two class encoding format.
• Vec Vector object.
• Vocabulary This object contains the list of words we want our recogniser to recognise plus a few "special" words (eg.
• WeightedMSECriterion Similar to MSECriterion, but you can put a weight on each example.
• WeightedSumMachine Weighted-sum machine.
• WordSegMeasurer This class can be used to save the word segmentation of a SimpleDecoderSpeechHMM in a file.
• XFile XFile.

Macros

• max The max function
• min The min function