altar.models.BayesianL2

Module Contents

Classes

class altar.models.BayesianL2.BayesianL2(name, locator, **kwds)

Bases: altar.models.Bayesian.Bayesian

A (Simplified) Bayesian Model with ParameterSets and L2 data norm

parameters
doc = the number of model degrees of freedom
cascaded
doc = whether the model is cascaded (annealing temperature is fixed at 1)
embedded
doc = whether the model is embedded in an ensemble of models
psets_list
doc = list of parameter sets, used to set orders
psets
default
doc = an ensemble of parameter sets in the model
dataobs
default
doc = observed data
case
doc = the directory with the input files
idx_map
default
doc = the indices for model parameters in whole theta set
return_residual
doc = the forward model returns residual(True) or prediction(False)
observations
device
precision
ifs
initialize(self, application)

Initialize the state of the model given an {application} context

posterior(self, application)

Sample my posterior distribution

initializeSample(self, step)

Fill {step.θ} with an initial random sample from my prior distribution.

verify(self, step, mask)

Check whether the samples in {step.theta} are consistent with the model requirements and update the {mask}, a vector with zeroes for valid samples and non-zero for invalid ones

evalPrior(self, theta, prior)

Fill {priorLLK} with the log likelihoods of the samples in {theta} in my prior distribution

abstract forwardModel(self, theta, prediction)

The forward model for a single set of parameters

forwardModelBatched(self, theta, prediction)

The forward model for a batch of theta: compute prediction from theta also return {residual}=True, False if the difference between data and prediction is computed

evalDataLikelihood(self, theta, likelihood)

calculate data likelihood and add it to step.prior or step.data

evalPosterior(self, step)

Given the {step.prior} and {step.data} likelihoods, compute a generalized posterior using {step.beta} and deposit the result in {step.post}

likelihoods(self, annealer, step)

Convenience function that computes all three likelihoods at once given the current {step} of the problem

updateModel(self, annealer)

Update Model parameters if needed :param annealer: :return: default is False

mountInputDataspace(self, pfs)

Mount the directory with my input files

loadFile(self, filename, shape=None, dataset=None, dtype=None)

Load an input file to a gsl vector or matrix (for both float32/64 support) Supported format: 1. text file in ‘.txt’ suffix, stored in prescribed shape 2. binary file with ‘.bin’ or ‘.dat’ suffix,

the precision must be same as the desired gpuprecision, and users must specify the shape of the data

  1. (preferred) hdf5 file in ‘.h5’ suffix (preferred)

    the metadata of shape, precision is included in .h5 file

Parameters:

  • filename – str, the input file name

  • shape – list of int

  • dataset – str, name/key of dataset for h5 input only

Returns:

output gsl vector/matrix

restrict(self, theta)

Return my portion of the sample matrix {theta}