Rename package to ExplainableAI (#44)

* Rename ExplainabilityMethods to ExplainableAI
* Generate new UUID using `UUIDs.uuid4()`
* Update readme

Author: adrhill

.github/workflows/ci.yml

@@ -60,8 +60,8 @@ jobs:
       - run: |
           julia --project=docs -e '
             using Documenter: doctest
-            using ExplainabilityMethods
-            doctest(ExplainabilityMethods)'
+            using ExplainableAI
+            doctest(ExplainableAI)'
       - run: julia --project=docs docs/make.jl
         env:
           DATADEPS_ALWAYS_ACCEPT: true # for MLDatasets download

Project.toml

@@ -1,7 +1,7 @@
-name = "ExplainabilityMethods"
-uuid = "cd722a4f-8d55-446b-8550-a4aabc9151ab"
+name = "ExplainableAI"
+uuid = "4f1bc3e1-d60d-4ed0-9367-9bdff9846d3b"
 authors = ["Adrian Hill"]
-version = "0.2.0"
+version = "0.3.0"
 
 [deps]
 ColorSchemes = "35d6a980-a343-548e-a6ea-1d62b119f2f4"

README.md

@@ -1,24 +1,28 @@
-![ExplainabilityMethods.jl][banner-img]
+![ExplainableAI.jl][banner-img]
 ___
 
+*Formerly known as ExplainabilityMethods.jl*
+
 | **Documentation**                                                     | **Build Status**                                      | **DOI**                 |
 |:--------------------------------------------------------------------- |:----------------------------------------------------- |:----------------------- |
 | [![][docs-stab-img]][docs-stab-url] [![][docs-dev-img]][docs-dev-url] | [![][ci-img]][ci-url] [![][codecov-img]][codecov-url] | [![][doi-img]][doi-url] |
 
 Explainable AI in Julia using [Flux.jl](https://fluxml.ai).
 
+This package implements interpretability methods and visualizations for neural networks, similar to [Captum](https://github.com/pytorch/captum) for PyTorch and [iNNvestigate](https://github.com/albermax/innvestigate) for Keras models. 
+
 ## Installation 
 To install this package and its dependencies, open the Julia REPL and run 
 ```julia-repl
-julia> ]add ExplainabilityMethods
+julia> ]add ExplainableAI
 ```
 
 ⚠️ This package is still in early development, expect breaking changes. ⚠️
 
 ## Example
 Let's use LRP to explain why an image of a cat gets classified as a cat:
 ```julia
-using ExplainabilityMethods
+using ExplainableAI
 using Flux
 using Metalhead
 
@@ -27,15 +31,14 @@ vgg = VGG19()
 model = strip_softmax(vgg.layers)
 
 # Run XAI method
-analyzer = LRPEpsilon(model)
+analyzer = LRP(model)
 expl = analyze(img, analyzer)
 
 # Show heatmap
 heatmap(expl)
 ```
 ![][heatmap]
 
-
 ## Methods
 Currently, the following analyzers are implemented:
 
@@ -48,29 +51,40 @@ Currently, the following analyzers are implemented:
     └── LRPGamma
 ```
 
-One of the design goals of ExplainabilityMethods.jl is extensibility.
+One of the design goals of ExplainableAI.jl is extensibility.
 Individual LRP rules like `ZeroRule`, `EpsilonRule`, `GammaRule` and `ZBoxRule` [can be composed][docs-composites] and are easily extended by [custom rules][docs-custom-rules].
 
+## Roadmap
+In the future, we would like to include:
+- [SmoothGrad](https://arxiv.org/abs/1706.03825)
+- [Integrated Gradients](https://arxiv.org/abs/1703.01365)
+- [PatternNet](https://arxiv.org/abs/1705.05598)
+- [DeepLift](https://arxiv.org/abs/1704.02685)
+- [LIME](https://arxiv.org/abs/1602.04938)
+- Shapley values via  [ShapML.jl](https://github.com/nredell/ShapML.jl)
+
+Contributions are welcome!
+
 ## Acknowledgements
 > Adrian Hill acknowledges support by the Federal Ministry of Education and Research (BMBF) for the Berlin Institute for the Foundations of Learning and Data (BIFOLD) (01IS18037A).
 
-[banner-img]: https://raw.githubusercontent.com/adrhill/ExplainabilityMethods.jl/gh-pages/assets/banner.png
-[heatmap]: https://raw.githubusercontent.com/adrhill/ExplainabilityMethods.jl/gh-pages/assets/heatmap.png
+[banner-img]: https://raw.githubusercontent.com/adrhill/ExplainableAI.jl/gh-pages/assets/banner.png
+[heatmap]: https://raw.githubusercontent.com/adrhill/ExplainableAI.jl/gh-pages/assets/heatmap.png
 
 [docs-stab-img]: https://img.shields.io/badge/docs-stable-blue.svg
-[docs-stab-url]: https://adrhill.github.io/ExplainabilityMethods.jl/stable
+[docs-stab-url]: https://adrhill.github.io/ExplainableAI.jl/stable
 
 [docs-dev-img]: https://img.shields.io/badge/docs-main-blue.svg
-[docs-dev-url]: https://adrhill.github.io/ExplainabilityMethods.jl/dev
+[docs-dev-url]: https://adrhill.github.io/ExplainableAI.jl/dev
 
-[ci-img]: https://github.com/adrhill/ExplainabilityMethods.jl/workflows/CI/badge.svg
-[ci-url]: https://github.com/adrhill/ExplainabilityMethods.jl/actions
+[ci-img]: https://github.com/adrhill/ExplainableAI.jl/workflows/CI/badge.svg
+[ci-url]: https://github.com/adrhill/ExplainableAI.jl/actions
 
-[codecov-img]: https://codecov.io/gh/adrhill/ExplainabilityMethods.jl/branch/master/graph/badge.svg
-[codecov-url]: https://codecov.io/gh/adrhill/ExplainabilityMethods.jl
+[codecov-img]: https://codecov.io/gh/adrhill/ExplainableAI.jl/branch/master/graph/badge.svg
+[codecov-url]: https://codecov.io/gh/adrhill/ExplainableAI.jl
 
-[docs-composites]: https://adrhill.github.io/ExplainabilityMethods.jl/dev/generated/advanced_lrp/#Custom-LRP-composites
-[docs-custom-rules]: https://adrhill.github.io/ExplainabilityMethods.jl/dev/generated/advanced_lrp/#Custom-LRP-rules
+[docs-composites]: https://adrhill.github.io/ExplainableAI.jl/dev/generated/advanced_lrp/#Custom-LRP-composites
+[docs-custom-rules]: https://adrhill.github.io/ExplainableAI.jl/dev/generated/advanced_lrp/#Custom-LRP-rules
 
 [doi-img]: https://zenodo.org/badge/337430397.svg
 [doi-url]: https://zenodo.org/badge/latestdoi/337430397

benchmark/Project.toml

@@ -1,5 +1,5 @@
 [deps]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
-ExplainabilityMethods = "cd722a4f-8d55-446b-8550-a4aabc9151ab"
+ExplainableAI = "4f1bc3e1-d60d-4ed0-9367-9bdff9846d3b"
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
 PkgBenchmark = "32113eaa-f34f-5b0d-bd6c-c81e245fc73d"

benchmark/benchmarks.jl

@@ -1,7 +1,7 @@
 using BenchmarkTools
 using Flux
-using ExplainabilityMethods
-import ExplainabilityMethods: modify_layer, lrp!
+using ExplainableAI
+import ExplainableAI: modify_layer, lrp!
 
 on_CI = haskey(ENV, "GITHUB_ACTIONS")
 

docs/Project.toml

@@ -2,7 +2,7 @@
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"
 ColorSchemes = "35d6a980-a343-548e-a6ea-1d62b119f2f4"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
-ExplainabilityMethods = "cd722a4f-8d55-446b-8550-a4aabc9151ab"
+ExplainableAI = "4f1bc3e1-d60d-4ed0-9367-9bdff9846d3b"
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
 ImageCore = "a09fc81d-aa75-5fe9-8630-4744c3626534"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"

docs/literate/advanced_lrp.jl

@@ -1,12 +1,12 @@
 # # Advanced LRP usage
-# One of the design goals of ExplainabilityMethods.jl is to combine ease of use and
+# One of the design goals of ExplainableAI.jl is to combine ease of use and
 # extensibility for the purpose of research.
 #
 #
 # This example will show you how to implement custom LRP rules and register custom layers
 # and activation functions.
 # For this purpose, we will quickly load our model from the previous section:
-using ExplainabilityMethods
+using ExplainableAI
 using Flux
 using MLDatasets
 using ImageCore
@@ -36,7 +36,7 @@ rules = [
 analyzer = LRP(model, rules)
 heatmap(input, analyzer)
 
-# Since some Flux Chains contain other Flux Chains, ExplainabilityMethods provides
+# Since some Flux Chains contain other Flux Chains, ExplainableAI provides
 # a utility function called [`flatten_model`](@ref).
 #
 #md # !!! warning "Flattening models"
@@ -51,7 +51,7 @@ struct MyGammaRule <: AbstractLRPRule end
 # It is then possible to dispatch on the utility functions  [`modify_params`](@ref) and [`modify_denominator`](@ref)
 # with the rule type `MyCustomLRPRule` to define custom rules without writing any boilerplate code.
 # To extend internal functions, import them explicitly:
-import ExplainabilityMethods: modify_params
+import ExplainableAI: modify_params
 
 function modify_params(::MyGammaRule, W, b)
     ρW = W + 0.25 * relu.(W)
@@ -68,7 +68,7 @@ heatmap(input, analyzer)
 analyzer = LRP(model, GammaRule())
 heatmap(input, analyzer)
 
-# If the layer doesn't use weights and biases `W` and `b`, ExplainabilityMethods provides a
+# If the layer doesn't use weights and biases `W` and `b`, ExplainableAI provides a
 # lower-level variant of [`modify_params`](@ref) called [`modify_layer`](@ref).
 # This function is expected to take a layer and return a new, modified layer.
 
@@ -111,22 +111,22 @@ model = Chain(model..., MyDoublingLayer())
 #   Layers failed model check
 #   ≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡
 #
-#   Found unknown layers MyDoublingLayer() that are not supported by ExplainabilityMethods' LRP implementation yet.
+#   Found unknown layers MyDoublingLayer() that are not supported by ExplainableAI's LRP implementation yet.
 #
-#   If you think the missing layer should be supported by default, please submit an issue (https://github.com/adrhill/ExplainabilityMethods.jl/issues).
+#   If you think the missing layer should be supported by default, please submit an issue (https://github.com/adrhill/ExplainableAI.jl/issues).
 #
 #   These model checks can be skipped at your own risk by setting the LRP-analyzer keyword argument skip_checks=true.
 #
 #   [...]
 # ```
 
-# LRP should only be used on deep rectifier networks and ExplainabilityMethods doesn't
+# LRP should only be used on deep rectifier networks and ExplainableAI doesn't
 # recognize `MyDoublingLayer` as a compatible layer.
 # By default, it will therefore return an error and a model check summary
 # instead of returning an incorrect explanation.
 #
 # However, if we know `MyDoublingLayer` is compatible with deep rectifier networks,
-# we can register it to tell ExplainabilityMethods that it is ok to use.
+# we can register it to tell ExplainableAI that it is ok to use.
 # This will be shown in the following section.
 
 #md # !!! warning "Skipping model checks"
@@ -136,7 +136,7 @@ model = Chain(model..., MyDoublingLayer())
 
 # ### Registering custom layers
 # The error in the model check will stop after registering our custom layer type
-# `MyDoublingLayer` as "supported" by ExplainabilityMethods.
+# `MyDoublingLayer` as "supported" by ExplainableAI.
 #
 # This is done using the function [`LRP_CONFIG.supports_layer`](@ref),
 # which should be set to return `true` for the type `MyDoublingLayer`:
@@ -175,7 +175,7 @@ model = Chain(flatten, Dense(784, 100, myrelu), Dense(100, 10))
 #
 #   Found layers with unknown or unsupported activation functions myrelu. LRP assumes that the model is a "deep rectifier network" that only contains ReLU-like activation functions.
 #
-#   If you think the missing activation function should be supported by default, please submit an issue (https://github.com/adrhill/ExplainabilityMethods.jl/issues).
+#   If you think the missing activation function should be supported by default, please submit an issue (https://github.com/adrhill/ExplainableAI.jl/issues).
 #
 #   These model checks can be skipped at your own risk by setting the LRP-analyzer keyword argument skip_checks=true.
 #
@@ -189,7 +189,7 @@ LRP_CONFIG.supports_activation(::typeof(myrelu)) = true
 analyzer = LRPZero(model)
 
 # ## How it works internally
-# Internally, ExplainabilityMethods dispatches to low level functions
+# Internally, ExplainableAI dispatches to low level functions
 # ```julia
 # function lrp!(rule, layer, Rₖ, aₖ, Rₖ₊₁)
 #     Rₖ .= ...
@@ -230,7 +230,7 @@ analyzer = LRPZero(model)
 #
 # and can be implemented via automatic differentiation (AD).
 #
-# This equation is implemented in ExplainabilityMethods as the default method
+# This equation is implemented in ExplainableAI as the default method
 # for all layer types that don't have a specialized implementation.
 # We will refer to it as the "AD fallback".
 #

docs/literate/example.jl

@@ -1,5 +1,5 @@
 # # Getting started
-# ExplainabilityMethods.jl can be used on any classifier.
+# ExplainableAI.jl can be used on any classifier.
 # In this first example, we will look at attributions on a LeNet5 model that was pretrained on MNIST.
 #
 # ### Loading the model
@@ -11,7 +11,7 @@
 #md #     @load "model.bson" model
 #md #     ```
 
-using ExplainabilityMethods
+using ExplainableAI
 using Flux
 using BSON
 
@@ -40,7 +40,7 @@ input = reshape(x, 28, 28, 1, :);
 
 #md # !!! warning "Input format"
 #md #
-#md #     For any attribution of a model, ExplainabilityMethods.jl assumes the batch dimension to be come last in the input.
+#md #     For any attribution of a model, ExplainableAI.jl assumes the batch dimension to be come last in the input.
 #md #
 #md #     For the purpose of heatmapping, the input is assumed to be in WHCN order
 #md #     (width, height, channels, batch), which is Flux.jl's convention.

docs/make.jl

@@ -1,4 +1,4 @@
-using ExplainabilityMethods
+using ExplainableAI
 using Documenter
 using Literate
 
@@ -15,9 +15,9 @@ for example in readdir(EXAMPLE_DIR)
 end
 
 makedocs(;
-    modules=[ExplainabilityMethods],
+    modules=[ExplainableAI],
     authors="Adrian Hill",
-    sitename="ExplainabilityMethods.jl",
+    sitename="ExplainableAI.jl",
     format=Documenter.HTML(; prettyurls=get(ENV, "CI", "false") == "true", assets=String[]),
     pages=[
         "Home" => "index.md",
@@ -27,4 +27,4 @@ makedocs(;
     ],
 )
 
-deploydocs(; repo="github.com/adrhill/ExplainabilityMethods.jl")
+deploydocs(; repo="github.com/adrhill/ExplainableAI.jl")

docs/src/api.md

@@ -1,5 +1,5 @@
 # Basics
-All methods in ExplainabilityMethods.jl work by calling `analyze` on an input and an analyzer:
+All methods in ExplainableAI.jl work by calling `analyze` on an input and an analyzer:
 ```@docs
 analyze
 heatmap