Change name from OMGP to AugmentedGaussianProcesses

Author: theogf

.travis.yml

@@ -11,5 +11,5 @@ notifications:
 after_success:
   # push coverage results to Coveralls
 - export DOCUMENTER_DEBUG=true
-- julia -e 'using Pkg; cd(Pkg.dir("OMGP")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder()); Codecov.submit(Codecov.process_folder())'
+- julia -e 'using Pkg; cd(Pkg.dir("AugmentedGaussianProcesses")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder()); Codecov.submit(Codecov.process_folder())'
 - julia -e 'using Pkg; ps=Pkg.PackageSpec(name="Documenter", version="0.19"); Pkg.add(ps); Pkg.pin(ps); include(joinpath("docs", "make.jl"))'

LICENSE.md

@@ -1,4 +1,4 @@
-The OMGP.jl package is licensed under the MIT "Expat" License:
+The AugmentedGaussianProcesses.jl package is licensed under the MIT "Expat" License:
 
 > Copyright (c) 2018: Théo Galy-Fajou.
 >

Project.toml

@@ -1,5 +1,5 @@
-name = "OMGP"
-uuid = "01d6f1c0-a789-11e8-1a9c-abf87115666d"
+name = "AugmentedGaussianProcesses"
+uuid = "b1c2f168-d918-11e8-38ee-334f3e5b0caa"
 authors = ["Theo Galy-Fajou <[email protected]>"]
 version = "0.3.1"
 

README.md

@@ -1,26 +1,37 @@
-# OMGP!
-[![Build Status](https://travis-ci.org/theogf/OMGP.jl.svg?branch=master)](https://travis-ci.org/theogf/OMGP.jl)
-  <a href="https://theogf.github.io/OMGP.jl/stable">
+# AugmentedGaussianProcesses!
+[![Build Status](https://travis-ci.org/theogf/AugmentedGaussianProcesses.jl.svg?branch=master)](https://travis-ci.org/theogf/AugmentedGaussianProcesses.jl)
+  <a href="https://theogf.github.io/AugmentedGaussianProcesses.jl/stable">
     <img src="https://img.shields.io/badge/docs-stable-blue.svg">
   </a>
-  <a href="https://theogf.github.io/OMGP.jl/latest">
+  <a href="https://theogf.github.io/AugmentedGaussianProcesses.jl/latest">
     <img src="https://img.shields.io/badge/docs-latest-blue.svg">
   </a>
-  
-Oh My GP! is a Julia package in development for **extremely efficient Gaussian Processes algorithms**. It contains for the moment two classifiers : one based on the Bayesian SVM (BSVM), and a state-of-the-art algorithm for classification using the logistic link called X-GPC (XGPC), two regression models one classic based on Gaussian likelihood (GPRegression) and another one in development base on Student T likelihood. A multiclass classifier model is in development.
+
+AugmentedGaussianProcesses! is a Julia package in development for **Data Augmented Gaussian Processes**. It contains a collection of models for different **gaussian and non-gaussian likelihoods**, which are transformed via data augmentation into **conditionally conjugate likelihood** allowing for **extremely fast inference** via block coordinate updates.
+
+# The package contains at the moment :
+  - Two GP classifier models
+    - **BSVM** : A Classifier with a likelihood equivalent to the classic SVM
+    - **XGPC** : A Classifier with a Bernoulli likelihood with the logistic link
+  - Two GP Regression models
+    - **GPRegression** : The standard Gaussian Process regression model with a Gaussian Likelihood (no data augmentation was needed here)
+    - **StudentT** : The standard Gaussian Process regression with a Student-t likelihood (the degree of freedom ν is not optimizable for the moment)
+  - More models in development
+    - **MultiClass** : A multiclass classifier model, relying on a modified version of softmax
+    - **Probit** : A Classifier with a Bernoulli likelihood with the probit link
+    - **Online** : Allowing for all algorithms to work online as well
 
 ## Install the package
 
 The package requires Julia 1.0
-Run in `Julia` press `]` and type `add OMGP` (once it has been released), it will install the package and all its requirements
-
+Run in `Julia` press `]` and type `add AugmentedGaussianProcesses` (once it has been released), it will install the package and all its requirements
 
 ## Use the package
 
 A complete documentation is currently being written, for now you can use this very basic example where `X_train` is a matrix ``N x D`` where `N` is the number of training points and `D` is the number of dimensions and `Y_train` is a vector of outputs.
 
 ```julia
-using OMGPC
+using AugmentedGaussianProcesses
 model = SparseXGPC(X_train,Y_train;Stochastic=true,batchsize=100,m=64,kernel=RBFKernel(1.0)) #Parameters after ; are optional
 model.train(iterations=100)
 Y_predic = model.predict(X_test) #For getting the label directly
@@ -40,6 +51,6 @@ Arxiv "Efficient Gaussian Process Classification using Polya-Gamma Variables" by
 UAI 13' "Gaussian Process for Big Data" by James Hensman, Nicolo Fusi and Neil D. Lawrence [https://arxiv.org/abs/1309.6835][arxivgpbigdata]
 
 [arxivgpbigdata]:https://arxiv.org/abs/1309.6835
-  [31b06e91]: https://github.com/theogf/OMGP.jl/blob/master/examples/Classification%20-%20SXGPC.ipynb "Classification with Sparse XGPC"
+  [31b06e91]: https://github.com/theogf/AugmentedGaussianProcesses.jl/blob/master/examples/Classification%20-%20SXGPC.ipynb "Classification with Sparse XGPC"
 [arxivbsvm]:https://arxiv.org/abs/1707.05532
 [arxivxgpc]:https://arxiv.org/abs/1802.06383

appveyor.yml

@@ -44,7 +44,7 @@ build_script:
 # Need to convert from shallow to complete for Pkg.clone to work
   - IF EXIST .git\shallow (git fetch --unshallow)
   - C:\projects\julia\bin\julia -e "versioninfo();
-      Pkg.clone(pwd(), \"OMGP\"); Pkg.build(\"OMGP\")"
+      Pkg.clone(pwd(), \"AugmentedGaussianProcesses\"); Pkg.build(\"AugmentedGaussianProcesses\")"
 
 test_script:
-  - C:\projects\julia\bin\julia -e "Pkg.test(\"OMGP\")"
+  - C:\projects\julia\bin\julia -e "Pkg.test(\"AugmentedGaussianProcesses\")"

docs/make.jl

@@ -1,9 +1,9 @@
-using Documenter, OMGP
+using Documenter, AugmentedGaussianProcesses
 
-makedocs(modules=[OMGP])
+makedocs(modules=[AugmentedGaussianProcesses])
 
 deploydocs(
     deps = Deps.pip("mkdocs", "python-markdown-math"),
-    repo = "github.com/theogf/OMGP.jl.git",
+    repo = "github.com/theogf/AugmentedGaussianProcesses.jl.git",
     julia = "1.0"
 )

docs/mkdocs.yml

@@ -1,8 +1,8 @@
-site_name:        OMGP.jl
-site_description: Website for OMGP.jl package
+site_name:        AugmentedGaussianProcesses.jl
+site_description: Website for AugmentedGaussianProcesses.jl package
 site_author:      Theo Galy-Fajou
-site_url:         https://github.com/theogf/OMGP.jl
-repo_url:         https://github.com/theogf/OMGP.jl
+site_url:         https://github.com/theogf/AugmentedGaussianProcesses.jl
+repo_url:         https://github.com/theogf/AugmentedGaussianProcesses.jl
 
 theme: readthedocs
 

docs/src/index.md

@@ -1,16 +1,16 @@
-# OMGP.jl Documentation
+# AugmentedGaussianProcesses.jl Documentation
 
 
 ```@contents
-OMGP
+AugmentedGaussianProcesses
 ```
 
 ## Index
 
 ```@index
-Modules = [OMGP]
+Modules = [AugmentedGaussianProcesses]
 ```
 
 # General introduction
 
-OMGP is a package providing
+AugmentedGaussianProcesses is a package providing

examples/Classification - SXGPC.ipynb

@@ -7,8 +7,8 @@
    "outputs": [],
    "source": [
     "using Plots; pyplot();\n",
-    "push!(LOAD_PATH,\"/home/theo/OMGP/src\")\n",
-    "using OMGP;"
+    "push!(LOAD_PATH,\"/home/theo/AugmentedGaussianProcesses/src\")\n",
+    "using AugmentedGaussianProcesses;"
    ]
   },
   {

examples/classification_SXGPC.jl

@@ -3,7 +3,7 @@ pyplot()
 #unicodeplots()
 using Distributions
 using ValueHistories
-using OMGP
+using AugmentedGaussianProcesses
 using ProfileView
 doPlot = false
 use_dataset = false
@@ -57,7 +57,7 @@ initPoints = copy(model.inducingPoints)
 iter_points = collect(1:1:1000)
 metrics = MVHistory()
 Parameters = MVHistory()
-function StoreIt(model::OMGP.GPModel,iter;hyper=false)#;iter_points=[],LogArrays=[],X_test=0,y_test=0)
+function StoreIt(model::AugmentedGaussianProcesses.GPModel,iter;hyper=false)#;iter_points=[],LogArrays=[],X_test=0,y_test=0)
     if in(iter,iter_points)
         if !hyper
             push!(metrics,:time_init,iter,time_ns()*1e-9)

perf/bsvm.jl

@@ -1,6 +1,6 @@
 cd(dirname(@__FILE__))
 using BenchmarkTools
-using OMGP
+using AugmentedGaussianProcesses
 using Distances, LinearAlgebra, Dates, MLDataUtils, DelimitedFiles
 suite = BenchmarkGroup()
 suite["Full"] = BenchmarkGroup(["init","elbo","computematrices","updatevariational","updatehyperparam","predic","predicproba"])
@@ -24,12 +24,12 @@ suite["Sparse"]["init"] = @benchmarkable SparseBSVM($X_train,$y_train,kernel=$ke
 suite["SparseStoch"]["init"] = @benchmarkable SparseBSVM($X_train,$y_train,kernel=$kernel,Autotuning=true,μ_init=ones(Float64,$m),Stochastic=true,m=$m,batchsize=$batchsize)
 for KT in ["Full","Sparse","SparseStoch"]
     models[KT].train(iterations=1)
-    suite[KT]["elbo"] = @benchmarkable OMGP.ELBO($(models[KT]))
-    suite[KT]["computematrices"] = @benchmarkable OMGP.computeMatrices!(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["updatevariational"] = @benchmarkable OMGP.variational_updates!(model,1) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["updatehyperparam"] = @benchmarkable OMGP.updateHyperParameters!(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["predic"] = @benchmarkable OMGP.probitpredict(model,$X_test) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["predicproba"] = @benchmarkable OMGP.probitpredictproba(model,X_test) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["elbo"] = @benchmarkable AugmentedGaussianProcesses.ELBO($(models[KT]))
+    suite[KT]["computematrices"] = @benchmarkable AugmentedGaussianProcesses.computeMatrices!(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["updatevariational"] = @benchmarkable AugmentedGaussianProcesses.variational_updates!(model,1) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["updatehyperparam"] = @benchmarkable AugmentedGaussianProcesses.updateHyperParameters!(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["predic"] = @benchmarkable AugmentedGaussianProcesses.probitpredict(model,$X_test) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["predicproba"] = @benchmarkable AugmentedGaussianProcesses.probitpredictproba(model,X_test) setup=(model=deepcopy($(models[KT])))
 end
 
 if isfile(paramfile)

perf/kernel.jl

@@ -1,6 +1,6 @@
 cd(dirname(@__FILE__))
 using BenchmarkTools,Traceur,Profile,ProfileView
-using OMGP.KernelModule
+using AugmentedGaussianProcesses.KernelModule
 using Distances, LinearAlgebra, Dates
 using Random: seed!
 suite = BenchmarkGroup()

perf/multiclass.jl

@@ -1,6 +1,6 @@
 cd(dirname(@__FILE__))
 using BenchmarkTools
-using OMGP
+using AugmentedGaussianProcesses
 using Distances, LinearAlgebra, Dates,  DelimitedFiles
 using MLDataUtils: splitobs
 suite = BenchmarkGroup()
@@ -33,12 +33,12 @@ suite["SparseStochKStoch"]["init"] = @benchmarkable SparseMultiClass($X_train,$y
 for KT in ["Full","FullKStoch","Sparse","SparseStoch","SparseStochKStoch"]
     println(KT)
     models[KT].train(iterations=1)
-    suite[KT]["elbo"] = @benchmarkable OMGP.ELBO(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["computematrices"] = @benchmarkable OMGP.computeMatrices!(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["updatevariational"] = @benchmarkable OMGP.variational_updates!(model,1) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["updatehyperparam"] = @benchmarkable OMGP.updateHyperParameters!(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["predic"] = @benchmarkable OMGP.multiclasspredict(model,$X_test) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["predicproba"] = @benchmarkable OMGP.multiclasspredictproba(model,$X_test) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["elbo"] = @benchmarkable AugmentedGaussianProcesses.ELBO(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["computematrices"] = @benchmarkable AugmentedGaussianProcesses.computeMatrices!(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["updatevariational"] = @benchmarkable AugmentedGaussianProcesses.variational_updates!(model,1) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["updatehyperparam"] = @benchmarkable AugmentedGaussianProcesses.updateHyperParameters!(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["predic"] = @benchmarkable AugmentedGaussianProcesses.multiclasspredict(model,$X_test) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["predicproba"] = @benchmarkable AugmentedGaussianProcesses.multiclasspredictproba(model,$X_test) setup=(model=deepcopy($(models[KT])))
 end
 
 if isfile(paramfile)

perf/xgpc.jl

@@ -1,6 +1,6 @@
 cd(dirname(@__FILE__))
 using BenchmarkTools
-using OMGP
+using AugmentedGaussianProcesses
 using Distances, LinearAlgebra, Dates, MLDataUtils, DelimitedFiles
 suite = BenchmarkGroup()
 suite["Full"] = BenchmarkGroup(["init","elbo","computematrices","updatevariational","updatehyperparam","predic","predicproba"])
@@ -24,12 +24,12 @@ suite["Sparse"]["init"] = @benchmarkable SparseXGPC($X_train,$y_train,kernel=$ke
 suite["SparseStoch"]["init"] = @benchmarkable SparseXGPC($X_train,$y_train,kernel=$kernel,Autotuning=true,μ_init=ones(Float64,$m),Stochastic=true,m=$m,batchsize=$batchsize)
 for KT in ["Full","Sparse","SparseStoch"]
     models[KT].train(iterations=1)
-    suite[KT]["elbo"] = @benchmarkable OMGP.ELBO($(models[KT]))
-    suite[KT]["computematrices"] = @benchmarkable OMGP.computeMatrices!(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["updatevariational"] = @benchmarkable OMGP.variational_updates!(model,1) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["updatehyperparam"] = @benchmarkable OMGP.updateHyperParameters!(model) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["predic"] = @benchmarkable OMGP.logitpredict(model,$X_test) setup=(model=deepcopy($(models[KT])))
-    suite[KT]["predicproba"] = @benchmarkable OMGP.logitpredictproba(model,X_test) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["elbo"] = @benchmarkable AugmentedGaussianProcesses.ELBO($(models[KT]))
+    suite[KT]["computematrices"] = @benchmarkable AugmentedGaussianProcesses.computeMatrices!(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["updatevariational"] = @benchmarkable AugmentedGaussianProcesses.variational_updates!(model,1) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["updatehyperparam"] = @benchmarkable AugmentedGaussianProcesses.updateHyperParameters!(model) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["predic"] = @benchmarkable AugmentedGaussianProcesses.logitpredict(model,$X_test) setup=(model=deepcopy($(models[KT])))
+    suite[KT]["predicproba"] = @benchmarkable AugmentedGaussianProcesses.logitpredictproba(model,X_test) setup=(model=deepcopy($(models[KT])))
 end
 
 if isfile(paramfile)

src/OMGP.jl

@@ -3,7 +3,7 @@
 General Framework for the data augmented Gaussian Processes
 
 """
-module OMGP
+module AugmentedGaussianProcesses
 
 @enum GPModelType Undefined=0 BSVM=1 XGPC=2 Regression=3 StudentT=4 MultiClassGP=5
 

src/functions/IO_model.jl

@@ -1,9 +1,9 @@
 # module IO_model
 
 using JLD2, FileIO
-# using OMGP
-import OMGP.KernelModule
-import OMGP.KernelModule.HyperParametersModule#: HyperParameter, OpenBound, NullBound, Interval
+# using AugmentedGaussianProcesses
+import AugmentedGaussianProcesses.KernelModule
+import AugmentedGaussianProcesses.KernelModule.HyperParametersModule#: HyperParameter, OpenBound, NullBound, Interval
 export save_trained_model, save_model,load_trained_model,load_model
 #Needs to be redefined
 function save_model(filename::String,model::GPModel)
@@ -127,15 +127,15 @@ end
 function writeextrabits(file,model::GPModel)
 end
 
-function create_model(sparse::Bool,modeltype::OMGP.GPModelType)
-    if modeltype == OMGP.GPModelType(1) #BSVM
+function create_model(sparse::Bool,modeltype::AugmentedGaussianProcesses.GPModelType)
+    if modeltype == AugmentedGaussianProcesses.GPModelType(1) #BSVM
         return model = sparse ? SparseBSVM() : BatchBSVM()
-    elseif modeltype == OMGP.GPModelType(2) #XGPC
+    elseif modeltype == AugmentedGaussianProcesses.GPModelType(2) #XGPC
         return model = sparse ? SparseXGPC() : BatchXGPC()
-    elseif modeltype == OMGP.GPModelType(3) #Regression
+    elseif modeltype == AugmentedGaussianProcesses.GPModelType(3) #Regression
         return model = sparse ? SparseGPRegression() :
         GPRegression()
-    elseif modeltype == OMGP.GPModelType(4) #StudentT
+    elseif modeltype == AugmentedGaussianProcesses.GPModelType(4) #StudentT
         return model = sparse ? SparseStudentT() : BatchStudentT()
     else
         @warn "Model writing for this model is not implemented yet"

src/kernels/hyperparameters/HyperParametersModule.jl

@@ -1,5 +1,5 @@
 module HyperParametersModule
-using OMGP.GradDescent
+using AugmentedGaussianProcesses.GradDescent
 import Base: convert, eltype, promote_type, show, string, ==, *, /, +, -, ^, isless;
 
 export

src/kmeans/KMeansModule.jl

@@ -7,7 +7,7 @@ using Distributions
 using StatsBase
 using LinearAlgebra
 using Clustering
-using OMGP.KernelModule
+using AugmentedGaussianProcesses.KernelModule
 
 export KMeansInducingPoints
 export KMeansAlg, StreamOnline, Webscale, CircleKMeans, DataSelection

src/models/StudentT_Functions.jl

@@ -1,7 +1,7 @@
 # Functions related to the Student T likelihood (StudentT)
 "Update the local variational parameters of the full batch GP StudentT"
 function local_update!(model::BatchStudentT)
-    model.β = 0.5.*(diag(model.Σ)+model.μ.^2-2.0.*model.μ.*model.y+(model.y.^2).+model.ν)
+    model.β = 0.5.*(diag(model.Σ)+(model.μ.-model.y).^2+model.ν)
     model.θ = 0.5.*(model.ν.+1.0)./model.β
 end
 
@@ -14,7 +14,7 @@ end
 
 "Update the local variational parameters of the sparse GP StudentT"
 function local_update!(model::SparseStudentT)
-    model.β = 0.5.*(model.Ktilde+sum((model.κ*model.Σ).*model.κ,dims=2)[:]+(model.κ*model.μ).^2-2.0.*(model.κ*model.μ).*model.y[model.MBIndices]+(model.y[model.MBIndices].^2).+model.ν)
+    model.β = 0.5.*(model.Ktilde+sum((model.κ*model.Σ).*model.κ,dims=2)[:]+(model.κ*model.μ.-model.y[model.MBIndices]).^2.+model.ν)
     model.θ = 0.5.*(model.ν.+1.0)./model.β
 
 end
@@ -92,8 +92,8 @@ end
 "Return the KL divergence for the inverse gamma distributions"
 function InverseGammaKL(model::GPModel)
     α_p = β_p = model.ν/2;
-    return (model.α-α_p)*digamma(model.α).-log(gamma(model.α)).+log(gamma(α_p))
-            .+ α_p.*(log.(model.β).-log(β_p)).+model.α.*(β_p.-model.β)./model.β
+    return (α_p.-model.α)*digamma(α_p).-log(gamma(α_p)).+log(gamma(model.α))
+            .+ model.α*(log(β_p).-log.(model.β)).+α_p.*(model.β.-β_p)./β_p
 end
 
 "Return a function computing the gradient of the ELBO given the kernel hyperparameters for a StudentT Model"

test/runtests.jl

@@ -1,4 +1,4 @@
-using OMGP
+using AugmentedGaussianProcesses
 using Test
 
 # write your own tests here