| Home >> All >> riso >> [ distributions Javadoc ] |
• riso.distributions: Javadoc index of package riso.distributions.
Package Samples:
• riso.distributions.computes_lambda_message
• riso.distributions.computes_pi_message
• riso.distributions.computes_pi
• riso.distributions.computes_lambda
• riso.distributions.computes_posterior
Classes:
• AbstractConditionalDistribution_AbstractDistribution_AbstractDistribution: This class implements a lambda message helper for a variable x with one or more parents u1,...,un . Except for the parent to which we are sending the lambda message, each parent sends a pi message to x . Let us suppose that we are sending the lambda message to parent uk . The lambda message is defined as p( e \ e_u1(above) | uk ) = \int p( x, e \ e_uk(above) | uk ) dx = \int p( x, e_x(below) + e_x(above) \ e_uk(above) | uk ) dx = \int p( e_x(below) | x ) p( x | e_x(above) \ e_uk(above), uk ) p( e_x(above) \ e_uk(above) | uk ) dx = p( e_x(above) \ e_uk(above) ) \int p( e_x(below) | x ) \int ... ...
• ConditionalGaussian: An instance of this class represents a conditional Gaussian distribution. The dependence enters only through the mean, which is a linear combination the parents plus an offset. The variance is constant. Writing the marginal means of the child and parent variables, respectively, as mu(1) and mu(2) , and the respective marginal variances as Sigma(11) and Sigma(22) , and the covariance as Sigma(12) , then the conditional mean mu(1|2) and conditional variance Sigma(1|2) are as follows. mu(1|2) = mu(1) + Sigma(12) Sigma(22)^{-1} (X(2)-mu(2)) Sigma(1|2) = Sigma(11) - Sigma(12) Sigma(22)^{-1} Sigma(21) ...
• ConditionalMixture: An instance of this class represents a conditional mixture model. This is similar to an unconditional mixture (represented by the Mixture class), but the mixing coefficients can vary with the context (i.e., the parent variables). In the Mixture class, mixing coefficients are stored in an array, since they don't change; here, each mixing coefficient is returned by a function that takes the context as an argument. In addition, the mixture components are conditional distributions, not unconditional. This class is declared abstract (i.e., it cannot be instantiated) because there is no generic way to ...
• AbstractDistribution: Abstract base class for unconditional distributions. Since Distribution is derived from ConditionalDistribution, any unconditional distribution must implement all of the functions defined for conditional distributions. Some of these have trivial implementations, which are given here. It would be cleaner, perhaps, to put these in the definition of Distribution, but Java doesn't allow code in an interface... . So here they are.
• AbstractConditionalDistribution: Abstract base class for conditional distributions. This class implements only a few methods; most of the methods from the ConditionalDistribution interface are not implemented, and so must be provided by subclasses. This classs is helpful in part because message-passing algorithms can be formulated as generic for all conditional distributions -- handlers are named only by classes, not by interfaces.
• Gaussian: A Gaussian (normal) distribution. The descriptive data which can be changed without causing the interface functions to break down is public. The other data is protected. Included in the public data are the regularization parameters. If not otherwise specified, the prior mean, prior covariance, and other regularization parameters are given neutral values, so that they have no effect on parameter estimation.
• MixGaussians: This class represents an additive mixture of Gaussian densities. There is little added functionality; the main thing is the name guarantees that all mixture components are Gaussian . The descriptive data which can be changed without causing the interface functions to break down is public. The other data is protected. Included in the public data are the regularization parameters.
• MixConditionalGaussians: An instance of this class represents a mixture of conditional Gaussian distributions. The dependence on the parents enters through the mean, which is assumed to be a linear combination of the parents plus an offset, and through the mixing coefficients. The variance of each component does not depend on the parents.
• MixGaussians: Yeah, it's sort of confusing that this class has the same name as in riso.distributions , but that's a consequence of the naming scheme used to locate message helpers. This class implements a helper which can handle a list of riso.distributions.MixGaussians messages.
• Gaussian: Yeah, it's sort of confusing that this class has the same name as in riso.distributions , but that's a consequence of the naming scheme used to locate message helpers. This class implements a helper which can handle a list of riso.distributions.Gaussian messages.
• Mixture: Yeah, it's sort of confusing that this class has the same name as in riso.distributions , but that's a consequence of the naming scheme used to locate message helpers. This class implements a helper which can handle a list of riso.distributions.Mixture messages.
• OuterProduct: An instance of this class represents an outer product of distributions, that is, a distribution which has a density of the form p(x) = \prod_i p_i(x[j_i]) where j_i is a subset of the indices 0,1,2,...,x.length-1 , with all subsets disjoint.
• AbstractConditionalDistribution_AbstractDistribution_: This class implements a lambda message helper for a variable x with one parents. The pi message from the parent is ignored; it should be null. This helper simply returns a helper from the more general helper class which handles variables with two or more parents.
• FunctionalRelation: This class is the superclass of all class which represent functional relations. A functional relation is a conditional distribution which has a conditional distribution which is a delta function. The location of the delta is determined by the f variable.
• Mixture: This class represents an additive mixture of distributions. The descriptive data which can be changed without causing the interface functions to break down is public. The other data is protected. Included in the public data are the regularization parameters.
• Truncated: An instance of this class represents a truncated distribution; the instance holds a reference to the distribution in question, and the effective support, mean value, etc., are computed from appropriate methods of the underlying distribution.
• Classifier: An instance of this class represents a classification model. Subclasses provide particular implementations of classification schemes. A classifier is the conditional distribution of a discrete child with discrete or continuous parents.
• IntegralCache: This class wraps the integral evaluation with a cache so that the integral need not be evaluated every time; if the integral has been evaluated for a nearby value of the special parent, an interpolated value is returned.
• SupportNotWellDefinedException: This exception is thrown when the effective support for a distribution cannot be computed because the effective support is not well-defined. Either the effective support is unbounded, or it is very large.
• GaussianDelta: An object of this class represents a continuous distribution whose mass is concentrated at one point. An instance of this class can be used anywhere that an instance of Gaussian can be used.
• DiscreteDelta: An object of this class represents a discrete distribution whose mass is concentrated at one point. An instance of this class can be used anywhere that an instance of Discrete can be used.
• Noninformative: An item of this class represents a ``noninformative'' lambda message or lambda function, that is, one for a variable which is not evidence and for which there is no downstream evidence.
• SquashingNetworkClassifier: An instance of this class represents a classification model based on a neural network. The neural network is set up to squash its outputs so that the outputs are in the range (0,1).
• LocationScaleDensity: Interface for so-called location and scale densities. These include the Gaussian, in which case the location is the mean and the scale is the covariance.
| Home | Contact Us | Privacy Policy | Terms of Service |