A utility package from the JuliaDiffusionBayes suite, used for defining observation schemes for stochastic processes. It is aimed primarily at encoding discrete-time observations of diffusions.
- Decorate each observation separately with the information about how it was collected
- Support for the following observations:
- Exact observations of all or a subset of all coordinates of the underlying process
- Linear translations of the underlying process, disturbed by Gaussian noise:
- First-passage time observations
- First-passage time observations with additional "resetting events"
- Non-linearly (or linearly) transformed observations with Gaussian or non-Gaussian noise, i.e.: with general function
gand random variable
- Parameterized versions of all observation types above
- Support for ergonomic definitions of
- Multiple observations of a single process
- Multiple observations of multiple processes, coming possibly from different laws that share subsets of parameters (mixed-effect models)
- Support for defining priors over starting points:
- Degenerate priors corresponding to fixed starting points
- Gaussian priors
] add ObservationSchemes
How To Use
See the documentation.
ObservationSchemes.jl belongs to a suite of packages in JuliaDiffusionBayes, whose aim is to facilitate Bayesian inference for diffusion processes. Some other packages in this suite are as follows:
- DiffusionDefinition.jl: define diffusion processes and sample from their laws
- GuidedProposals.jl: defining and sampling conditioned diffusion processes
- ExtensibleMCMC.jl: a modular implementation of the Markov chain Monte Carlo (MCMC) algorithms
- DiffusionMCMC.jl: Markov chain Monte Carlo (MCMC) algorithms for doing inference for diffusion processes