fix include SubSampleObjective as part of ParamSpaceSGD

Author: Red-Portal

src/algorithms/paramspacesgd/interface.jl

@@ -1,11 +1,11 @@
 
 """
-This family of algorithms (`<:KLMinRepGradDescent`,`<:KLMinRepGradProxDescent`,`<:KLMinScoreGradDescent`) applies stochastic gradient descent (SGD) to the variational `objective` over the (Euclidean) space of variational parameters.
+This family of algorithms (`<:KLMinRepGradDescent`,`<:KLMinRepGradProxDescent`,`<:KLMinScoreGradDescent`,`<:SubsampledObjective`) applies stochastic gradient descent (SGD) to the variational `objective` over the (Euclidean) space of variational parameters.
 The trainable parameters in the variational approximation are expected to be extractable through `Optimisers.destructure`.
 This requires the variational approximation to be marked as a functor through `Functors.@functor`.
 """
 const ParamSpaceSGD = Union{
-    <:KLMinRepGradDescent,<:KLMinRepGradProxDescent,<:KLMinScoreGradDescent
+    <:KLMinRepGradDescent,<:KLMinRepGradProxDescent,<:KLMinScoreGradDescent,<:SubsampledObjective
 }
 
 function init(rng::Random.AbstractRNG, alg::ParamSpaceSGD, q_init, prob)