Add serialization how_to

Author: limarta

docs/pages.jl

@@ -24,6 +24,7 @@ pages = [
         "Custom Modeling Languages" => "how_to/custom_dsl.md",
         "Custom Gradients" => "how_to/custom_derivatives.md",
         "Incremental Computation" => "how_to/custom_incremental_computation.md",
+        "Serializing Dynamic DSL Traces" => "how_to/serialization.md"
     ],
     "API Reference" => [
         "Modeling Library" => [

docs/src/how_to/serialization.md

@@ -0,0 +1,31 @@
+# [Saving Traces between REPL Sessions](@id how_to_serialize)
+
+When saving a trace to disk, it is likely to run into issues using `Serialization.jl` since traces internally track function pointers (recall `Trace`s may contain function pointers). Instead, you can try the experiemental [`GenSerialization.jl`](https://github.com/probcomp/GenSerialization.jl) which uses can discard function call data. This is most useful for check-pointing work (e.g. inference) between different REPL sessions on the same machine. 
+
+Since `Serialization.jl` is used underneath, similar restrictions apply (see [`serialize`](https://docs.julialang.org/en/v1/stdlib/Serialization/#Serialization.serialize)). Please not we do not guarantee portability between different machines and have yet extensively test this.
+
+!!! note
+    See the repo for warnings and limitations.
+
+
+An example:
+
+```julia
+using Gen
+using GenSerialization
+
+@gen function model(p) 
+    x ~ bernoulli(p)
+end
+
+trace = simulate(model, (0.2))
+serialize("coin_flip.gen", trace)
+```
+
+This stores the trace in a file. Now to deserialize, run
+```julia
+saved_trace = deserialize("coin_flip.gen", model)
+```
+!!! note
+    The same generative function used to save out the trace must be defined at runtime before deserialization.
+