Home

Calibrate-Emulate-Sample

Purpose

Construct a registered (?) package for a black-box (approximate) uncertainty quantification for noisy, expensive and non-differentiable models.

Short term

• Refactor the code into the structure detailed below
• Interface with EnsembleKalmanProcesses.jl

Longer term

• Working with user model instability. (We could also do nothing here, or we could)
• Public Availability?
• Example library from CliMA

5 Latest features / Developments

1. (PR #83) EKS bugfix, now runs with runtests.

To do's

Functionality

Overall Code structure

Mind-map form of project.

Data Structures

Here we include the data structures we use in the project

ParameterDistribution

ParameterDistribution(...)

Module contains the additional functions

set_distribution()
get_distribution()
sample_distribution()
transform_real_to_dist(), 
transform_dist_to_real()
apply_units_to_real()

ProcessedData Storage

ProcessedData(...)

Module contains the additional functions

set_data()
get_data()
data_mean()
data_cov()

Calibrate: Interface

Here we include the interface with the EnsembleKalmanProcesses.jl module.

EnsembleKalmanProcessRuns(...)

Workflow 1

ModelInterface(...)

Workflow 2

ModelInterface(...)

Emulate: Objects

GPEmulator(...)

Sample:

MCMC(...)

Visualization:

This will be performed through the vizCES.jl module

Docs(?)