Cover Problem

CoverParams(; [sel_tax=0.0], [set_shape=1.0],
              [min_weight=1E-2], [mask_discount=0.9], [setXset_factor=1.0],
              [uncovered_factor=0.1], [covered_factor=0.001],
              [set_relevance_shape=0.5], [set_relevance_min=0.5]) -> CoverParams

Specify parameters for the AbstractCoverProblem. One can specify a non-default value for a particular parameter.

See cover quality score for the detailed description of the parameters.

Arguments

• sel_tax: $\epsilon$, the constant added to the set score of each selected set to penalize the number of sets in the cover
• set_shape: $\beta$, applied to set or set×set scores
• min_weight: minimal non-zero set probability ???
• mask_discount: $\alpha$, how much the overlap score of each subsequent mask (from most to less enriched) is discounted
• setXset_factor: $w_r$, the weight of redundancy cover quality component (setXset_score scale), 0 = no redunancy penalty
• uncovered_factor: $w_u$, the weight of uncovered hits component in cover quality
• covered_factor: $w_c$, the weight of covered non-hits component in cover quality
• set_relevance_shape: $\beta_L$, how much set relevance affects set score, 0 = no effect
• set_relevance_min: $L_{\min$, if shaped relevance is below, it's set to set_relevance_min

Linear component of an individual masked set score for the CoverProblem. Doesn't take into account the overlap with the other selected sets.

Parameters for collect(mosaic::MaskedSetMosaic).

AbstractCoverProblem{T}

Optimal Enriched-Set Cover problem – choose the sets from the collection 𝒞 to cover the masked(selected) elements M. The optimal sets cover C = {c₁, c₂, ..., cₙ} ⊂ 𝒞 has to deliver 3 goals:

• be relevant (i.e. minimize the P-values of M and cᵢ sets overlap)
• be minimal (i.e. minimize the number of sets in C)
• be non-redundant (i.e. minimize the P-values of the pairwise non-overlap of C sets with each other).

Fuzzy set selection is possible – each set is assigned a weight from [0, 1] range.

Total number of problem variables.

Abstract multi-objective optimal Enriched-Set Cover problem.

See "Method Description" for more details.

Multi-objective optimal Enriched-Set Cover problem that uses MaskedSetMosaic to define the coverage scores based on the set weights and masked elements coverage.

See "Method Description" for more details.

Parameters for the Borg-based optimization of MultiobjCoverProblem.

See optimize.

optimize(problem::MultiobjCoverProblem,
         [opt_params::MultiobjOptimizerParams]) -> MultiobjCoverProblemResult

Optimize MultiobjCoverProblem and return the result. Uses Borf multi-objective optimization method from BlackBoxOptim.jl package.

Transforms the 4-component cover quality score into 2-component score that makes the highly redundant solutions dominated by any less redundant ones.

See "Cover score convolution" section for the discussion.

Arguments

• setXset_factor: $w_r$, same as in CoverParams
• uncovered_factor: $w_u$, same as in CoverParams
• covered_factor: $w_c$, same as in CoverParams
• ratio_threshold: $k_{\max}$, defaults to 1
• shape: $\alpha_k$, defaults to 0.5.

score(w::AbstractVector{Float64}, problem) -> NTuple{4, Float64}

Unfolded multiobjective score (fitness) of the OESC coverage.

• w: probabilities of the sets being covered

See "Cover quality".

The result of optimize. Contains the solutions on the Pareto front: weights of the annotation terms and corresponding cover scores.

The collection of masked set covers.

Greedy enumeration of enriched-set covers.

• At each iteration an optimal enriched-set cover problem is being solved.
• The sets selected at current iteration are removed from further consideration.
• The process continues with the reduced collection until the result is an empty collection or the last cover is much worse than the first one.

Returns CoverCollection.

Convert covers, a collection of the covers of mosaic, into a DataFrame.