Model

Creates a model object containing a log likelihood function and prior distributions

• prior: a vector of distribution objects for each parameter. A uniform prior is used if no

prior is passed.

• loglike: a function that computes the log likelihood

Constructor

Model(args...; prior=nothing, loglike, kwargs...)

Examples

# Default uninform prior
loglike(θ, design, data) = ...

model = Model(;loglike)

# Custom priors
prior = [Beta(5,5),...]
loglike(θ, design, data) = ...

model = Model(;loglike, prior)

Optimizer(;task, model, grid_design, grid_parms, grid_response)

Optimizer constructs a model object for adaptive design optimization

Fields

• model: a model object
• grid_design:a grid of design parameters
• grid_parms: a grid of model parameters
• grid_data: a grid of data
• log_like: a three dimensional array of precomputed log likelihoods
• marg_log_like: a two dimensional array containing marginal log likelihoods for design and data
• priors: a multidimensional array of prior probabilities for parameters
• log_post: a one dimensional array of log posterior probabilities for parameters
• entropy: a two dimensional array of entropy values for parameter and design combinations
• marg_entropy:
• cond_entropy:
• mutual_info:
• best_design:
• parm_names:
• model_state:
• update_state!:

loglikelihood(optimizer::Optimizer)

Computes a three dimensional grid of log likelihoods where the first dimension is the model parameters, the second dimension is the design parameters, and the third dimension is the data values.

• loglike: a model object containing a log likelihood function and prior distributions
• parm_grid: a grid of model parameters
• design_grid: a grid of design parameters
• data_grid: a grid of data values
• model_type: model type can be Static or Dynamic
• model_state: model state variables that are updated for Dynamic models

loglikelihood(optimizer::Optimizer)

Computes a three dimensional grid of log likelihoods for a dynamic model where the first dimension is the model parameters, the second dimension is the design parameters, and the third dimension is the data values.

• optimizer: an optimizer object

loglikelihood(model::Model, design_grid, parm_grid, data_grid, model_type, model_state)

Computes a three dimensional grid of log likelihoods where the first dimension is the model parameters, the second dimension is the design parameters, and the third dimension is the data values.

• loglike: a model object containing a log likelihood function and prior distributions
• parm_grid: a grid of model parameters
• design_grid: a grid of design parameters
• data_grid: a grid of data values
• model_type: model type can be Static or Dynamic
• model_state: model state variables that are updated for Dynamic models

loglikelihood(model::Model, design_grid, parm_grid, data_grid, model_type, model_state)

Computes a three dimensional grid of log likelihoods for a dynamic model where the first dimension is the model parameters, the second dimension is the design parameters, and the third dimension is the data values.

• loglike: a model object containing a log likelihood function and prior distributions
• parm_grid: a grid of model parameters
• design_grid: a grid of design parameters
• data_grid: a grid of data values
• model_type: model type can be Static or Dynamic
• model_state: model state variables that are updated for Dynamic models

loglikelihood(model::Model, design_grid, parm_grid, data_grid, model_type, model_state)

Computes a three dimensional grid of log likelihoods where the first dimension is the model parameters, the second dimension is the design parameters, and the third dimension is the data values.

• model::Model: a model object containing a log likelihood function and prior distributions
• parm_grid: a grid of model parameters
• design_grid: a grid of design parameters
• data_grid: a grid of data values
• model_type: model type can be Static or Dynamic
• model_state: model state variables that are updated for Dynamic models

loglikelihood(model::Model, design_grid, parm_grid, data_grid, model_type, model_state)

Computes a three dimensional grid of log likelihoods where the first dimension is the model parameters, the second dimension is the design parameters, and the third dimension is the data values.

• model::Model: a model object containing a log likelihood function and prior distributions
• parm_grid: a grid of model parameters
• design_grid: a grid of design parameters
• data_grid: a grid of data values
• model_type: model type can be Static or Dynamic
• model_state: model state variables that are updated for Dynamic models

marginal_log_like!(optimizer)

Marginalizes the log likelihoods over model parameters, resulting in a three dimensional array where the first dimension is length 1, the second dimension is the design parameters, and the third dimension is the data values.

• optimizer: an optimizer object

marginal_log_like(optimizer)

Marginalizes the log likelihoods over model parameters, resulting in a three dimensional array where the first dimension is length 1, the second dimension is the design parameters, and the third dimension is the data values.

• log_post: a vector of posterior log likelihood values for the model parameters
• log_like: a three dimensional vector of log likelihood values where the first dimension corresponds to the model parameters,

the second dimension corresponds to the design parameters, and the third dimension corresponds to the data values

mutual_information!(optimizer)

Computes a vector of the mutual information values where each element corresponds to a design parameter.

• optimizer: an optimizer object

mutual_information(optimizer)

Computes a vector of the mutual information values where each element corresponds to a design parameter.

• marg_entropy: a vector of marginal entropy values where each element corresponds to a design parameter
• cond_entropy: a vector of conditional entropy values where each element corresponds to a design parameter

prior_probs(prior, parm_grid)

Computes a grid of prior probabilities over model parameters.

• prior: a vector of Distribution objects, one for each parameter
• parm_grid: a grid of parameter values

prior_probs(model::Model, parm_grid)

Computes a grid of prior probabilities over model parameters.

• model::Model: a model object
• parm_grid: a grid of parameter values

prior_probs(prior, parm_grid)

Computes a grid of uniform prior probabilities over model parameters.

• prior::Nothing: a vector of Distribution objects, one for each parameter
• parm_grid: a grid of parameter values