Allow Gradient analyzers on non-Flux models (#150)

* Allow Gradient analyzers on non-Flux models

* Fix typo in `BATCHDIM_MISSING` error

Author: adrhill

src/analyze_api.jl

@@ -4,7 +4,7 @@ abstract type AbstractXAIMethod end
 
 const BATCHDIM_MISSING = ArgumentError(
     """The input is a 1D vector and therefore missing the required batch dimension.
-    Call `analyze` with the keyword argument `add_batch_dim=false`."""
+    Call analyze with the keyword argument add_batch_dim=true."""
 )
 
 """
@@ -46,16 +46,14 @@ end
 
 # lower-level call to method
 function _analyze(
-    input::AbstractArray{T,N},
+    input::AbstractArray,
     method::AbstractXAIMethod,
     sel::AbstractNeuronSelector;
     add_batch_dim::Bool=false,
     kwargs...,
-) where {T<:Real,N}
-    if add_batch_dim
-        return method(batch_dim_view(input), sel; kwargs...)
-    end
-    N < 2 && throw(BATCHDIM_MISSING)
+)
+    add_batch_dim && (input = batch_dim_view(input))
+    ndims(input) < 2 && throw(BATCHDIM_MISSING)
     return method(input, sel; kwargs...)
 end
 

src/gradient.jl

@@ -14,10 +14,12 @@ end
 
 Analyze model by calculating the gradient of a neuron activation with respect to the input.
 """
-struct Gradient{C<:Chain} <: AbstractXAIMethod
-    model::C
+struct Gradient{M} <: AbstractXAIMethod
+    model::M
+    Gradient(model) = new{typeof(model)}(model)
     Gradient(model::Chain) = new{typeof(model)}(Flux.testmode!(check_output_softmax(model)))
 end
+
 function (analyzer::Gradient)(input, ns::AbstractNeuronSelector)
     grad, output, output_indices = gradient_wrt_input(analyzer.model, input, ns)
     return Explanation(grad, output, output_indices, :Gradient, nothing)
@@ -29,12 +31,14 @@ end
 Analyze model by calculating the gradient of a neuron activation with respect to the input.
 This gradient is then multiplied element-wise with the input.
 """
-struct InputTimesGradient{C<:Chain} <: AbstractXAIMethod
-    model::C
+struct InputTimesGradient{M} <: AbstractXAIMethod
+    model::M
+    InputTimesGradient(model) = new{typeof(model)}(model)
     function InputTimesGradient(model::Chain)
-        return new{typeof(model)}(Flux.testmode!(check_output_softmax(model)))
+        new{typeof(model)}(Flux.testmode!(check_output_softmax(model)))
     end
 end
+
 function (analyzer::InputTimesGradient)(input, ns::AbstractNeuronSelector)
     grad, output, output_indices = gradient_wrt_input(analyzer.model, input, ns)
     attr = input .* grad

src/heatmap.jl

@@ -42,15 +42,15 @@ See also [`analyze`](@ref).
     instead of computing it individually for each sample. Defaults to `false`.
 """
 function heatmap(
-    val::AbstractArray{T,N};
+    val::AbstractArray;
     cs::ColorScheme=ColorSchemes.seismic,
     reduce::Symbol=:sum,
     rangescale::Symbol=:centered,
     permute::Bool=true,
     unpack_singleton::Bool=true,
     process_batch::Bool=false,
-) where {T,N}
-    N != 4 && throw(
+)
+    ndims(val) != 4 && throw(
         ArgumentError(
             "heatmap assumes Flux's WHCN convention (width, height, color channels, batch size) for the input.
             Please reshape your explanation to match this format if your model doesn't adhere to this convention.",