Update XAIBase to v3.0.0 (#13)

* Update XAIBase to v3.0.0 * Update `AbstractNeuronSelector` to `AbstractOutputSelector` * Update README to load VisionHeatmaps * Load VisionHeatmaps in docs * Add `LayerNormRule` to rule overview
Julia-XAI · Feb 21, 2024 · d4857bb · d4857bb
1 parent 56127f0
commit d4857bb
Show file tree

Hide file tree

Showing 11 changed files with 30 additions and 21 deletions.
diff --git a/Project.toml b/Project.toml
@@ -1,7 +1,7 @@
 name = "RelevancePropagation"
 uuid = "0be6dd02-ae9e-43eb-b318-c6e81d6890d8"
 authors = ["Adrian Hill <gh@adrianhill.de>"]
-version = "1.1.0"
+version = "2.0.0-DEV"
 
 [deps]
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
@@ -20,6 +20,6 @@ Markdown = "1"
 Random = "1"
 Reexport = "1"
 Statistics = "1"
-XAIBase = "1.3"
+XAIBase = "3"
 Zygote = "0.6"
 julia = "1.6"
diff --git a/README.md b/README.md
@@ -25,6 +25,7 @@ using a pre-trained VGG16 model from [Metalhead.jl](https://github.com/FluxML/Me
 
 ```julia
 using RelevancePropagation
+using VisionHeatmaps             # visualization of explanations as heatmaps
 using Flux
 using Metalhead                  # pre-trained vision models
 
@@ -40,7 +41,7 @@ input = ...                      # input in WHCN format
 composite = EpsilonPlusFlat()
 analyzer = LRP(model, composite)
 expl = analyze(input, analyzer)  # or: expl = analyzer(input)
-heatmap(expl)                    # Show heatmap
+heatmap(expl)                    # show heatmap using VisionHeatmaps.jl
 
 ```
 
@@ -64,8 +65,6 @@ whereas regions in blue are of negative relevance.
 | `LRP` with `EpsilonGammaBox` composite        | ![][castle-lrp-egb]            | ![][streetsign-lrp-egb]            |
 | `LRP`                                         | ![][castle-lrp]                | ![][streetsign-lrp]                |
 
-
-
 ## 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).

diff --git a/docs/Project.toml b/docs/Project.toml
@@ -10,3 +10,7 @@ ImageShow = "4e3cecfd-b093-5904-9786-8bbb286a6a31"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
 MLDatasets = "eb30cadb-4394-5ae3-aed4-317e484a6458"
 RelevancePropagation = "0be6dd02-ae9e-43eb-b318-c6e81d6890d8"
+VisionHeatmaps = "27106da1-f8bc-4ca8-8c66-9b8289f1e035"
+
+[compat]
+VisionHeatmaps = "1.4"
diff --git a/docs/src/api.md b/docs/src/api.md
@@ -4,9 +4,14 @@ All methods in RelevancePropagation.jl work by calling `analyze` on an input and
 ```@docs
 analyze
 Explanation
-heatmap
 ```
 
+For heatmapping functionality, take a look at either
+[VisionHeatmaps.jl](https://julia-xai.github.io/XAIDocs/VisionHeatmaps/stable/) or
+[TextHeatmaps.jl](https://julia-xai.github.io/XAIDocs/TextHeatmaps/stable/).
+Both provide `heatmap` methods for visualizing explanations, 
+either for images or text, respectively.
+
 ## LRP analyzer
 ```@docs
 LRP

diff --git a/docs/src/literate/basics.jl b/docs/src/literate/basics.jl
@@ -100,7 +100,12 @@ convert2image(MNIST, x)
 analyzer = LRP(model)
 
 # This ana lyzer will return heatmaps that look identical to the `InputTimesGradient` analyzer
-# from [ExplainableAI.jl](https://github.com/Julia-XAI/ExplainableAI.jl):
+# from [ExplainableAI.jl](https://github.com/Julia-XAI/ExplainableAI.jl).
+# We can visualize `Explanation`s by computing a `heatmap` using either
+# [VisionHeatmaps.jl](https://julia-xai.github.io/XAIDocs/VisionHeatmaps/stable/) or
+# [TextHeatmaps.jl](https://julia-xai.github.io/XAIDocs/TextHeatmaps/stable/),
+# either for images or text, respectively.
+using VisionHeatmaps
 
 heatmap(input, analyzer)
 

diff --git a/docs/src/literate/crp.jl b/docs/src/literate/crp.jl
@@ -46,6 +46,8 @@ features = IndexedFeatures(1, 2, 10)
 # ## Step 3: Use CRP analyzer
 # We can now create a [`CRP`](@ref) analyzer
 # and use it like any other analyzer from RelevancePropagation.jl:
+using VisionHeatmaps
+
 analyzer = CRP(lrp_analyzer, feature_layer, features)
 heatmap(input, analyzer)
 

diff --git a/docs/src/literate/custom_rules.jl b/docs/src/literate/custom_rules.jl
@@ -10,6 +10,7 @@
 
 # We start out by loading the same pre-trained LeNet-5 model and MNIST input data:
 using RelevancePropagation
+using VisionHeatmaps
 using Flux
 using MLDatasets
 using ImageCore
@@ -63,7 +64,8 @@ rules = [
     ZeroRule(),
 ]
 analyzer = LRP(model, rules)
-heatmap(input, analyzer)
+
+heatmap(input, analyzer) # using VisionHeatmaps.jl
 
 # We just implemented our own version of the ``γ``-rule in 2 lines of code.
 # The heatmap perfectly matches the pre-implemented `GammaRule`:

diff --git a/docs/src/rules.md b/docs/src/rules.md
@@ -44,6 +44,7 @@ ZBoxRule
 ```
 
 ## Specialized rules
-```@docs; canonical=false 
+```@docs; canonical=false
+LayerNormRule
 GeneralizedGammaRule
 ```
diff --git a/src/crp.jl b/src/crp.jl
@@ -32,7 +32,7 @@ end
 # Call to CRP analyzer #
 #======================#
 
-function (crp::CRP)(input::AbstractArray{T,N}, ns::AbstractNeuronSelector) where {T,N}
+function (crp::CRP)(input::AbstractArray{T,N}, ns::AbstractOutputSelector) where {T,N}
     rules = crp.lrp.rules
     layers = crp.lrp.model.layers
     modified_layers = crp.lrp.modified_layers

diff --git a/src/lrp.jl b/src/lrp.jl
@@ -56,7 +56,7 @@ LRP(model::Chain, c::Composite; kwargs...) = LRP(model, lrp_rules(model, c); kwa
 #==========================#
 
 function (lrp::LRP)(
-    input::AbstractArray, ns::AbstractNeuronSelector; layerwise_relevances=false
+    input::AbstractArray, ns::AbstractOutputSelector; layerwise_relevances=false
 )
     as = get_activations(lrp.model, input)    # compute activations aᵏ for all layers k
     Rs = similar.(as)                         # allocate relevances Rᵏ for all layers k
@@ -69,7 +69,7 @@ end
 
 get_activations(model, input) = (input, Flux.activations(model, input)...)
 
-function mask_output_neuron!(R_out, a_out, ns::AbstractNeuronSelector)
+function mask_output_neuron!(R_out, a_out, ns::AbstractOutputSelector)
     fill!(R_out, 0)
     idx = ns(a_out)
     R_out[idx] .= 1

diff --git a/test/Project.toml b/test/Project.toml
@@ -12,12 +12,3 @@ Suppressor = "fd094767-a336-5f1f-9728-57cf17d0bbfb"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 XAIBase = "9b48221d-a747-4c1b-9860-46a1d8ba24a7"
 
-[compat]
-Aqua = "0.8"
-Flux = "0.13, 0.14"
-JLD2 = "0.4"
-Metalhead = "0.8 - 0.9.1"
-NNlib = "0.8, 0.9"
-ReferenceTests = "0.10"
-Suppressor = "0.2"
-XAIBase = "1.2"