Merge branch 'main' into dev

.github/workflows/ci.yml

@@ -0,0 +1,33 @@
+name: Run tests
+
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - master
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        julia-version: ['1.8']
+        julia-arch: [x64]
+        os: [ubuntu-latest]
+
+    steps:
+      - uses: actions/checkout@v2
+      - uses: julia-actions/[email protected]
+        with:
+          version: ${{ matrix.julia-version }}
+      - uses: julia-actions/julia-buildpkg@v1
+      - uses: julia-actions/julia-runtest@v1
+
+      - uses: julia-actions/julia-processcoverage@v1
+
+      - uses: codecov/codecov-action@v3
+        with:
+          files: ./coverage1.xml,./coverage2.xml,./lcov.info # optional
+          flags: unittests # optional
+          name: codecov-umbrella # optional

README.md

@@ -1,28 +1,28 @@
+[![Run tests](https://github.com/aicenter/SumProductSet.jl/actions/workflows/ci.yml/badge.svg)](https://github.com/aicenter/SumProductSet.jl/actions/workflows/ci.yml)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 # SumProductSet.jl
 
-is package for probailistic learning of hierarchicaly structured heterogeneous data based on SumProduct networks. The package extends standard SumProduct networks by introducing new type of model node and implementing rules for building models over aforementioned data. It is based on [SumProductTransform.jl](https://github.com/pevnak/SumProductTransform.jl) package. The package [Mill.jl](https://github.com/CTUAvastLab/Mill.jl) is used as a framework that unifies accepted data format.
-
-
+is a package for probabilistic learning of tree-structured, heterogeneous graph data based on sum-product networks. The package extends the standard sum-product networks by introducing a novel computational node---the set node---and creating model-builders for the networks. The implementation follows the methodology of the [SumProductTransform.jl](https://github.com/pevnak/SumProductTransform.jl) package. The package [Mill.jl](https://github.com/CTUAvastLab/Mill.jl) is utilized to process tree-structured graphs embodied by the JSON format.
 
 To reproduce this project, do the following:
 
-1. Download this code base repository.
-2. Open a Julia (preferably Julia 1.8) console and type:
+1. Download this repository.
+2. Open the Julia (preferably Julia 1.8) console and type:
    ```julia
    using Pkg
    Pkg.activate("path/to/the/project")
    Pkg.instantiate()
    ```
-   This will download and install all necessary packages defined in project environment for you.
+   These commands will download and install all necessary packages defined in the project environment.
 3. Use the package:
    ``` julia
    using SumProductSet
 
    ```
 
-   Basic examples are shown in [examples](https://github.com/aicenter/SumProductSet.jl/tree/dev/examples) folder. Examples that require externel libraries
-   not included in SumProductSet.jl package have separate environment.
+   Basic examples are in the [examples](https://github.com/aicenter/SumProductSet.jl/tree/dev/examples) folder. Examples requiring libraries that are
+   not included in SumProductSet.jl package have a separate environment.
 
    ## Note
-   Importing Mill.jl or Distribution.jl is prefered to `using` these packages since it does not create ambiguity between the packages and SumProductSet.jl
+   Adopting `import` Mill.jl or Distribution.jl is preferred to `using`, since it does not create ambiguity between these packages and the SumProductSet.jl
    package.

examples/base_spsn/set_classification.jl

@@ -29,14 +29,14 @@ end
 function knownsetmixture(μs, Σs, λs, prior)
     components = map(zip(μs, Σs, λs)) do ps
         pc = Poisson(log(ps[3]))
-        pf = MvNormalParams(ps[1], ps[2])
+        pf = MvNormalParams(Float32.(ps[1]), Float32.(ps[2]))
         SetNode(pf, pc)
     end
     SumNode(components, prior)    
 end
 
 ### Creating and sampling known model
-m1 = knownsetmixture(μ, Σ, λ, [1., 1., 1.])
+m1 = knownsetmixture(μ, Σ, λ, [1f0, 1, 1])
 nbags = 100
 bags, baglabels = randwithlabel(m1, nbags)
 

examples/base_spsn/set_clustering.jl

@@ -31,13 +31,13 @@ end
 function knownsetmixture(μs, Σs, λs, prior)
     components = map(zip(μs, Σs, λs)) do ps
         pc = Poisson(log(ps[3]))
-        pf = MvNormalParams(ps[1], ps[2])
+        pf = MvNormalParams(Float32.(ps[1]), Float32.(ps[2]))
         SetNode(pf, pc)
     end
     SumNode(components, prior)    
 end
 ### Creating and sampling known model
-m1 = knownsetmixture(μ, Σ, λ, [1., 1., 1.])
+m1 = knownsetmixture(μ, Σ, λ, [1f0, 1, 1])
 nbags = 300
 bags, baglabels = randwithlabel(m1, nbags)
 

examples/mutagenesis/base_model.ipynb

@@ -436,4 +436,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 5
-}
+}

src/distributions/distributions.jl

@@ -10,6 +10,7 @@ include("categorical.jl")
 include("geometric.jl")
 include("bernoulli.jl")
 
+
 ####
 #	Functions for making the library compatible with Mill
 ####

src/distributions/geometric.jl

@@ -29,7 +29,7 @@ julia> logpdf(m, x)
 
 
 struct Geometric{T} <: Distribution
-    logitp::Array{T, 1}
+    logitp::Vector{T}
 end
 
 Flux.@functor Geometric

src/distributions/mvnormal.jl

@@ -34,6 +34,7 @@ struct MvNormal{T, N} <: Distribution
 end
 
 Flux.@functor MvNormal
+Flux.trainable(m::MvNormal) = (m.b, m.A,)
 
 MvNormal{T, N}(b::Array{T, 1}, A::Array{T, N}) where {T, N} = MvNormal(b, A, zero(T))
 

test/distributions/mvnormal.jl

@@ -45,7 +45,7 @@ end
     end
 end
 
-@testset "MvNormal correctness" begin
+@testset "MvNormal --- correctness" begin
     μ = [-3., 11]
     Σ = [5. 3; 3 7]
     n = 10000