using Revise, SossGen
import Gen
m = @model σ begin
μ ~ Normal()
y ~ Normal(μ, σ) |> iid(100000)
end;
args = (σ=1.1,);
truth = rand(m(args));
pairs(truth)
pairs(::NamedTuple) with 3 entries:
:σ => 1.1
:μ => 0.398088
:y => [-0.849513, 0.777053, -0.800264, 0.105648, -1.50177, 1.26158, -0.43
2465…
g1 = SossGenerativeFunction(m);
g2 = SossGenerativeFunction(m,codegen);
# tr = simulate(g, (args,)) # generate a full trace
tr1, w1 = generate(g1, (args,), choicemap(:y => truth.y)); # generate a trace with y = 3
tr2, w2 = generate(g2, (args,), choicemap(:y => truth.y)); # generate a trace with y = 3
@gen (static) function q(current, args)
@trace(normal(current[:μ], 0.2), :μ)
end
Gen.load_generated_functions();
using BenchmarkTools
@btime mh($tr1, $q, $(args,));
891.681 μs (147 allocations: 8.64 KiB)
@btime mh($tr2, $q, $(args,));
24.929 μs (148 allocations: 8.67 KiB)