Interfaces

Templates

The template describe the image features we wish to extract. As of 0.11 VIDA uses the ComradeBase.jl and VLBISkyModels.jl interface. This means that any model that obeys that interface can be used within VIDA. Additionally VIDA defines a number of additional templates that are useful. For a complete list see the API page.

For specifically VIDA you should subtype VLBISkyModels.AbstractImageTemplate subtype of the ComradeBase.AbstractModel type and partially implemented some functions. For instance, we assume that ComradeBase.imanalytic(::Type{<:AbstractImageTemplate}) = IsAnalytic(). As such if an end user wants to implement a new feature they just need to implement

ComradeBase.intensity_point(m::MyNewTemplate, p)
ComradeBase.radialextent(m::MyNewTemplate)

Note we do not implement the flux of the templates since they are often difficult to calculate.

Divergence AbstractDivergence

In order to extract a feature you need to create a probability divergence function. Currently the divergences are defined using a VIDA.AbstractDivergence type. The general user-facing interface is

bh = Bhattacharyya(image) #make the Bh divergence
kl = KullbackLeibler(image) #makes the KL divergence

to initialize the divergence. To evaluate the divergence on a template you use the VIDA.divergence function

divergence(bh, θ::AbstractTemplate)
divergence(kl, θ::AbstractTemplate)

Extract features vida and VIDAProblem

The main goal of VIDA is to extract image features. To do this we need to define the template and parameterization we want to use. The first step is to create a template function that takes in a NamedTuple and returns an <:ComradeBase.AbstractModel. For example

temp(θ) = GaussianRing(θ.r0, θ.σ, θ.x0, θ.y0)

For our search we also need to provide the domain over which we want to search

lower = (r0 = 5.0, σ = 0.1, x0 = -60.0, y0 = 60.0)
upper = (r0 = 30.0, σ = 5.0, x0 = -60.0, y0 = 60.0)

We can then form our VIDA.VIDAProblem using the divergence defined above

prob = VIDAProblem(bh, temp, lower, upper)

Finally, to get the optimal parameters and template we can call the VIDA.vida function

using OptimizationMetaheuristics
xopt, opt_template, divmin = vida(prob, ECA())