MPIRecoPlans

In this explanation we focus on the blueprint mechanism RecoPlans of AbstractImageReconstruction and how these are used in MPIReco. We start with a quick recap of the RecoPlan interface and then see how MPIReco uses it.

RecoPlans

RecoPlans are blueprints from which parameters and algorithms can be constructed. They are essentially thin-wrappers around nested-key value pairs. Plans are able to represent the nested tree structure of algorithms and their parameters. A plan can either be empty, partially or fully parameterized. Empty plans are helpful to describe just the structure of an algorithm, which a user can then parameterize. If a plan is constructed this way, it is missing all parameters:

using MPIReco.AbstractImageReconstruction
plan = RecoPlan(CommonPreProcessingParameters)
plan.frames

missing

We can interact with the plan as if it were a mutable variant of the given parameters:

plan.frames = 1:10
plan.bgParams = NoBackgroundCorrectionParameters()
plan

RecoPlan{CommonPreProcessingParameters}
├─ tfCorrection
├─ neglectBGFrames
├─ numPeriodAverages
├─ numPeriodGrouping
├─ numAverages
├─ spectralLeakageCorrection
├─ bgParams = NoBackgroundCorrectionParameters()
├─ frames = 1:10
└─ loadasreal

And we can construct instances of parameters and algorithms from a plan:

parameter = build(plan)

CommonPreProcessingParameters{NoBackgroundCorrectionParameters}(NoBackgroundCorrectionParameters(), 1:10, nothing, true, 1, 1, 1, false, false)

Likewise we can go from an instance to a plan via:

plan = toPlan(parameter)

RecoPlan{CommonPreProcessingParameters}
├─ tfCorrection = nothing
├─ neglectBGFrames = true
├─ numPeriodAverages = 1
├─ numPeriodGrouping = 1
├─ numAverages = 1
├─ spectralLeakageCorrection = false
├─ bgParams::RecoPlan{NoBackgroundCorrectionParameters}
├─ frames = 1:10
└─ loadasreal = false

We can either manually empty a plan from its parameters:

plan.frames = missing

missing

Or use the clear! method, which preserves the structure of a blueprint by default:

clear!(plan)

RecoPlan{CommonPreProcessingParameters}
├─ tfCorrection
├─ neglectBGFrames
├─ numPeriodAverages
├─ numPeriodGrouping
├─ numAverages
├─ spectralLeakageCorrection
├─ bgParams::RecoPlan{NoBackgroundCorrectionParameters}
├─ frames
└─ loadasreal

RecoPlans can be written to and read from files:

plan.frames = 1:42
plan.loadasreal = true
plan.numAverages = rand(1:100)
toTOML(joinpath(@__DIR__, "Parameters.toml"), plan)

loadasreal = true
numAverages = 44
_module = "MPIReco"
_type = "RecoPlan{CommonPreProcessingParameters}"

[frames]
_unionmodule = "Core"
_module = "Core"
_type = "Union"
_uniontype = "UnitRange{Int64}"

    [frames._value]
    stop = 42
    start = 1
    _module = "Base"
    _type = "UnitRange"

[bgParams]
_module = "MPIReco"
_type = "RecoPlan{NoBackgroundCorrectionParameters}"

This way one can either serialize a completely parametrized image reconstruction or just the blueprint of an algorithm.

For more information on RecoPlans, we refer to the How-to section of the AbstractImageReconstruction documentation.

MPIRecoPlan

MPIReco facilitates access to RecoPlans via two additional features. The first features deals with loading RecoPlans. To load a RecoPlan from a file one has to provide the filename and a list of modules in which the respective algorithms and parameters are defined. MPIReco tracks a list of directories and modules to grant a user easier access to plans contained within the directories.

As was shown in the Implement Reconstruction Packages how-to, it is possible to register new modules and directory to MPIReco:

addRecoPlanPath(@__DIR__())
plan = MPIRecoPlan(joinpath(@__DIR__, "Parameters"))

RecoPlan{CommonPreProcessingParameters}
├─ tfCorrection
├─ neglectBGFrames
├─ numPeriodAverages
├─ numPeriodGrouping
├─ numAverages = 44
├─ spectralLeakageCorrection
├─ bgParams::RecoPlan{NoBackgroundCorrectionParameters}
├─ frames = 1:42
└─ loadasreal = true

The second feature was shown in the Enable Caching how-to. This feature is realized by MPIReco keeping a least-recently-used (LRU) cache for recently opened plans. As long as the plan structure was not changed in the file or the parameters, a plan can be reused from the cache. While this also saves costs in loading a plan, the main benefit is reusing the same plan instance. If the plan in question contains processing steps that implement caching, all algorithms derived from a plan can access this cache. This enables an algorithm to reuse the results of previous processing steps.

This page was generated using Literate.jl.