Add Basic Structure (#3)

* Basic Structure

* Update Project.toml

* Address review suggestions

* Modify julia compat

* Update Project.toml

* Add likelihood

* Add test and fix typo

* Add trainable

* Update .travis.yml

* Remove logpdf for likelihood and fix params and trainable issue.

* CHange TravisCI documentation's required Julia version to 1

* Add Poisson Likelihood

* Fix bug

* Remove lambda parameter for Poisson likelihood

* Add randf and randy and remove trainable default

* Rename \Phi to lik

* Move different likelihoods into directories

* Remove custom rand functions

* Use Functors, remove Requires and Flux dependency

* Minor style fixes

* Remove Likelihood type

* Address code review by @devmotion

* Setup test/Project.toml

Author: sharanry

.github/workflows/CompatHelper.yml

@@ -0,0 +1,16 @@
+name: CompatHelper
+
+on:
+  schedule:
+    - cron: '00 00 * * *'
+
+jobs:
+  CompatHelper:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Pkg.add("CompatHelper")
+        run: julia -e 'using Pkg; Pkg.add("CompatHelper")'
+      - name: CompatHelper.main()
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: julia -e 'using CompatHelper; CompatHelper.main(; subdirs = ["", "test", "docs"])'

.travis.yml

@@ -4,7 +4,7 @@ os:
   - linux
   - osx
 julia:
-  - 1.0
+  - 1
   - nightly
 notifications:
   email: false
@@ -16,7 +16,7 @@ jobs:
   fast_finish: true
   include:
     - stage: Documentation
-      julia: 1.0
+      julia: 1
       script: julia --project=docs -e '
           using Pkg;
           Pkg.develop(PackageSpec(path=pwd()));

Project.toml

@@ -3,11 +3,18 @@ uuid = "45b822f7-b0f9-4a40-9473-dffa0412f8e7"
 authors = ["willtebbutt <[email protected]>"]
 version = "0.1.0"
 
-[compat]
-julia = "1"
-
-[extras]
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+[deps]
+AbstractGPs = "99985d1d-32ba-4be9-9821-2ec096f28918"
+Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
+Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
+KernelFunctions = "ec8451be-7e33-11e9-00cf-bbf324bd1392"
+LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
-[targets]
-test = ["Test"]
+[compat]
+AbstractGPs = "0.1.1"
+Distributions = "0.19, 0.20, 0.21, 0.22, 0.23"
+Functors = "0.1"
+KernelFunctions = "0.4"
+julia = "1.3"

src/LatentGPs.jl

@@ -1,5 +1,26 @@
 module LatentGPs
 
-greet() = print("Hello World!")
+using Distributions
+using KernelFunctions
+using AbstractGPs
+using LinearAlgebra
+using Random
+using Functors
+
+import Statistics
+import Distributions
+
+export LatentGP
+
+export logpdf, rand
+
+export GaussianLikelihood, PoissonLikelihood
+
+
+include("latent_gp.jl")
+
+# Likelihoods
+include("likelihoods/gaussian.jl")
+include("likelihoods/poisson.jl")
 
 end # module

src/latent_gp.jl

@@ -0,0 +1,30 @@
+"""
+    LatentGP(f<:GP, lik)
+
+ - `fx` is a `FiniteGP`.
+ - `lik` is the log likelihood function which maps sample from f to corresposing 
+ conditional likelihood distributions.
+    
+"""
+struct LatentGP{T<:AbstractGPs.FiniteGP, S}
+    fx::T
+    lik::S
+end
+
+function Distributions.rand(rng::AbstractRNG, lgp::LatentGP)
+    f = rand(rng, lgp.fx)
+    y = rand(rng, lgp.lik(f))
+    return (f=f, y=y)
+end
+
+"""
+    logpdf(lgp::LatentGP, y::NamedTuple{(:f, :y)})
+
+```math
+    log p(y, f; x)
+```
+Returns the joint log density of the gaussian process output `f` and real output `y`.
+"""
+function Distributions.logpdf(lgp::LatentGP, y::NamedTuple{(:f, :y)})
+    return logpdf(lgp.fx, y.f) + logpdf(lgp.lik(y.f), y.y)
+end

src/likelihoods/gaussian.jl

@@ -0,0 +1,22 @@
+"""
+    GaussianLikelihood(σ²)
+
+Gaussian likelihood with `σ²` variance. This is to be used if we assume that the 
+uncertainity associated with the data follows a Gaussian distribution.
+
+```math
+    p(y|f) = Normal(y | f, σ²)
+```
+On calling, this would return a normal distribution with mean `f` and variance σ².
+"""
+struct GaussianLikelihood{T<:Real}
+    σ²::T
+end
+
+GaussianLikelihood() = GaussianLikelihood(1e-6)
+
+@functor GaussianLikelihood
+
+(l::GaussianLikelihood)(f::Real) = Normal(f, sqrt(l.σ²))
+
+(l::GaussianLikelihood)(fs::AbstractVector{<:Real}) = MvNormal(fs, sqrt(l.σ²))

src/likelihoods/poisson.jl

@@ -0,0 +1,11 @@
+"""
+    PoissonLikelihood()
+
+Poisson likelihood with rate as exponential of samples from GP `f`. This is to be used if
+we assume that the uncertainity associated with the data follows a Poisson distribution.
+"""
+struct PoissonLikelihood end
+
+(l::PoissonLikelihood)(f::Real) = Poisson(exp(f))
+
+(l::PoissonLikelihood)(fs::AbstractVector{<:Real}) = Product(Poisson.(exp.(fs)))

test/Project.toml

@@ -0,0 +1,14 @@
+[deps]
+AbstractGPs = "99985d1d-32ba-4be9-9821-2ec096f28918"
+Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
+Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
+KernelFunctions = "ec8451be-7e33-11e9-00cf-bbf324bd1392"
+LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+
+[compat]
+AbstractGPs = "0.1.1"
+Distributions = "0.19, 0.20, 0.21, 0.22, 0.23"
+KernelFunctions = "0.4"
+julia = "1.3"

test/latent_gp.jl

@@ -0,0 +1,15 @@
+@testset "latent_gp" begin
+    gp = GP(SqExponentialKernel())
+    x = rand(10)
+    y = rand(10)
+    fx = gp(x, 1e-5)
+
+    lik = GaussianLikelihood(1e-5)
+    
+    lgp = LatentGP(fx, lik)
+    @test typeof(lgp) <: LatentGP
+    @test typeof(lgp.fx) <: AbstractGPs.FiniteGP
+    f = rand(10)
+    @test typeof(logpdf(lgp, (f=f, y=y))) <: Real
+    @test typeof(rand(lgp)) <: NamedTuple{(:f, :y)}
+end

test/likelihoods/gaussian.jl

@@ -0,0 +1,12 @@
+@testset "GaussianLikelihood" begin
+    gp = GP(SqExponentialKernel())
+    x = rand(10)
+    y = rand(10)
+    fx = gp(x, 1e-5)
+    lik = GaussianLikelihood(first(fx.Σy))
+    lgp = LatentGP(fx, lik)
+    
+    @test typeof(lik(rand(fx))) <: Distribution
+    @test length(rand(lik(rand(fx)))) == 10
+    @test keys(Functors.functor(lik)[1]) == (:σ²,)
+end