Document LRP composites (#43)

Author: adrhill

README.md

@@ -69,8 +69,8 @@ Individual LRP rules like `ZeroRule`, `EpsilonRule`, `GammaRule` and `ZBoxRule`
 [codecov-img]: https://codecov.io/gh/adrhill/ExplainabilityMethods.jl/branch/master/graph/badge.svg
 [codecov-url]: https://codecov.io/gh/adrhill/ExplainabilityMethods.jl
 
-[docs-composites]: https://adrhill.github.io/ExplainabilityMethods.jl/dev/generated/example/#Custom-composites
-[docs-custom-rules]: https://adrhill.github.io/ExplainabilityMethods.jl/dev/generated/example/#Custom-rules
+[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
 
 [doi-img]: https://zenodo.org/badge/337430397.svg
 [doi-url]: https://zenodo.org/badge/latestdoi/337430397

docs/literate/advanced_lrp.jl

@@ -18,6 +18,31 @@ index = 10
 x, y = MNIST.testdata(Float32, index)
 input = reshape(x, 28, 28, 1, :);
 
+# ## Custom LRP composites
+# Instead of creating an LRP-analyzer from a single rule (e.g. `LRP(model, GammaRule())`),
+# we can also assign rules to each layer individually.
+# For this purpose, we create an array of rules that matches the length of the Flux chain:
+rules = [
+    ZBoxRule(),
+    GammaRule(),
+    GammaRule(),
+    EpsilonRule(),
+    EpsilonRule(),
+    EpsilonRule(),
+    ZeroRule(),
+    ZeroRule(),
+]
+
+analyzer = LRP(model, rules)
+heatmap(input, analyzer)
+
+# Since some Flux Chains contain other Flux Chains, ExplainabilityMethods provides
+# a utility function called [`flatten_model`](@ref).
+#
+#md # !!! warning "Flattening models"
+#md #     Not all models can be flattened, e.g. those using
+#md #     `Parallel` and `SkipConnection` layers.
+
 # ## Custom LRP rules
 # Let's define a rule that modifies the weights and biases of our layer on the forward pass.
 # The rule has to be of type `AbstractLRPRule`.