Merge pull request #334 from femtomc/20201116_mrb_switch_combinator

(Ready for review): Switch combinator

Author: marcoct

docs/src/ref/combinators.md

@@ -119,4 +119,41 @@ TODO: document me
     <img src="../../images/recurse_combinator.png" alt="schematic of recurse combinatokr" width="70%"/>
 </div>
 ```
+## Switch combinator
 
+```@docs
+Switch
+```
+
+In the schematic below, the kernel is denoted `S` and accepts an integer index `k`.
+
+Consider the following constructions:
+
+```julia
+@gen function bang((grad)(x::Float64), (grad)(y::Float64))
+    std::Float64 = 3.0
+    z = @trace(normal(x + y, std), :z)
+    return z
+end
+
+@gen function fuzz((grad)(x::Float64), (grad)(y::Float64))
+    std::Float64 = 3.0
+    z = @trace(normal(x + 2 * y, std), :z)
+    return z
+end
+
+sc = Switch(bang, fuzz)
+```
+
+This creates a new generative function `sc`. We can then obtain the trace of `sc`:
+
+```julia
+(trace, _) = simulate(sc, (2, 5.0, 3.0))
+```
+
+The resulting trace contains the subtrace from the branch with index `2` - in this case, a call to `fuzz`:
+
+```
+│
+└── :z : 13.552870875213735
+```

src/modeling_library/cond.jl

@@ -0,0 +1,20 @@
+# ------------ Switch trace ------------ #
+
+struct SwitchTrace{T} <: Trace
+    gen_fn::GenerativeFunction{T}
+    index::Int
+    branch::Trace
+    retval::T
+    args::Tuple
+    score::Float64
+    noise::Float64
+end
+
+@inline get_choices(tr::SwitchTrace) = get_choices(tr.branch)
+@inline get_retval(tr::SwitchTrace) = tr.retval
+@inline get_args(tr::SwitchTrace) = tr.args
+@inline get_score(tr::SwitchTrace) = tr.score
+@inline get_gen_fn(tr::SwitchTrace) = tr.gen_fn
+@inline Base.getindex(tr::SwitchTrace, addr) = Base.getindex(tr.branch, addr)
+@inline project(tr::SwitchTrace, selection::Selection) = project(tr.branch, selection)
+@inline project(tr::SwitchTrace, ::EmptySelection) = tr.noise

src/modeling_library/modeling_library.jl

@@ -66,12 +66,16 @@ include("dist_dsl/dist_dsl.jl")
 # code shared by vector-shaped combinators
 include("vector.jl")
 
+# traces for with prob/switch combinator
+include("cond.jl")
+
 # built-in generative function combinators
 include("choice_at/choice_at.jl")
 include("call_at/call_at.jl")
 include("map/map.jl")
 include("unfold/unfold.jl")
 include("recurse/recurse.jl")
+include("switch/switch.jl")
 
 #############################################################
 # abstractions for constructing custom generative functions #

src/modeling_library/switch/assess.jl

@@ -0,0 +1,26 @@
+mutable struct SwitchAssessState{T}
+    weight::Float64
+    retval::T
+    SwitchAssessState{T}(weight::Float64) where T = new{T}(weight)
+end
+
+function process!(gen_fn::Switch{C, N, K, T},
+                  index::Int, 
+                  args::Tuple, 
+                  choices::ChoiceMap, 
+                  state::SwitchAssessState{T}) where {C, N, K, T}
+    (weight, retval) = assess(getindex(gen_fn.branches, index), args, choices)
+    state.weight = weight
+    state.retval = retval
+end
+
+@inline process!(gen_fn::Switch{C, N, K, T}, index::C, args::Tuple, choices::ChoiceMap, state::SwitchAssessState{T}) where {C, N, K, T} = process!(gen_fn, getindex(gen_fn.cases, index), args, choices, state)
+
+function assess(gen_fn::Switch{C, N, K, T}, 
+                args::Tuple, 
+                choices::ChoiceMap) where {C, N, K, T}
+    index = args[1]
+    state = SwitchAssessState{T}(0.0)
+    process!(gen_fn, index, args[2 : end], choices, state)
+    return state.weight, state.retval
+end

src/modeling_library/switch/backprop.jl

@@ -0,0 +1,2 @@
+@inline choice_gradients(trace::SwitchTrace{T}, selection::Selection, retval_grad) where T = choice_gradients(getfield(trace, :branch), selection, retval_grad)
+@inline accumulate_param_gradients!(trace::SwitchTrace{T}, retval_grad, scale_factor = 1.) where {T} = accumulate_param_gradients!(getfield(trace, :branch), retval_grad, scale_factor)

src/modeling_library/switch/generate.jl

@@ -0,0 +1,34 @@
+mutable struct SwitchGenerateState{T}
+    score::Float64
+    noise::Float64
+    weight::Float64
+    index::Int
+    subtrace::Trace
+    retval::T
+    SwitchGenerateState{T}(score::Float64, noise::Float64, weight::Float64) where T = new{T}(score, noise, weight)
+end
+
+function process!(gen_fn::Switch{C, N, K, T},
+                  index::Int, 
+                  args::Tuple, 
+                  choices::ChoiceMap, 
+                  state::SwitchGenerateState{T}) where {C, N, K, T}
+
+    (subtrace, weight) = generate(getindex(gen_fn.branches, index), args, choices)
+    state.index = index
+    state.subtrace = subtrace
+    state.weight += weight
+    state.retval = get_retval(subtrace)
+end
+
+@inline process!(gen_fn::Switch{C, N, K, T}, index::C, args::Tuple, choices::ChoiceMap, state::SwitchGenerateState{T}) where {C, N, K, T} = process!(gen_fn, getindex(gen_fn.cases, index), args, choices, state)
+
+function generate(gen_fn::Switch{C, N, K, T}, 
+                  args::Tuple, 
+                  choices::ChoiceMap) where {C, N, K, T}
+
+    index = args[1]
+    state = SwitchGenerateState{T}(0.0, 0.0, 0.0)
+    process!(gen_fn, index, args[2 : end], choices, state)
+    return SwitchTrace{T}(gen_fn, state.index, state.subtrace, state.retval, args[2 : end], state.score, state.noise), state.weight
+end

src/modeling_library/switch/propose.jl

@@ -0,0 +1,29 @@
+mutable struct SwitchProposeState{T}
+    choices::DynamicChoiceMap
+    weight::Float64
+    retval::T
+    SwitchProposeState{T}(choices, weight) where T = new{T}(choices, weight)
+end
+
+function process!(gen_fn::Switch{C, N, K, T}, 
+                  index::Int, 
+                  args::Tuple, 
+                  state::SwitchProposeState{T}) where {C, N, K, T}
+
+    (submap, weight, retval) = propose(getindex(gen_fn.branches, index), args)
+    state.choices = submap
+    state.weight += weight
+    state.retval = retval
+end
+
+@inline process!(gen_fn::Switch{C, N, K, T}, index::C, args::Tuple, state::SwitchProposeState{T}) where {C, N, K, T} = process!(gen_fn, getindex(gen_fn.cases, index), args, state)
+
+function propose(gen_fn::Switch{C, N, K, T}, 
+                 args::Tuple) where {C, N, K, T}
+
+    index = args[1]
+    choices = choicemap()
+    state = SwitchProposeState{T}(choices, 0.0)
+    process!(gen_fn, index, args[2:end], state)
+    return state.choices, state.weight, state.retval
+end

src/modeling_library/switch/regenerate.jl

@@ -0,0 +1,60 @@
+mutable struct SwitchRegenerateState{T}
+    weight::Float64
+    score::Float64
+    noise::Float64
+    prev_trace::Trace
+    trace::Trace
+    index::Int
+    retdiff::Diff
+    SwitchRegenerateState{T}(weight::Float64, score::Float64, noise::Float64, prev_trace::Trace) where T = new{T}(weight, score, noise, prev_trace)
+end
+
+function process!(gen_fn::Switch{C, N, K, T},
+                  index::Int,
+                  index_argdiff::Diff,
+                  args::Tuple,
+                  kernel_argdiffs::Tuple,
+                  selection::Selection, 
+                  state::SwitchRegenerateState{T}) where {C, N, K, T}
+    branch_fn = getfield(gen_fn.branches, index)
+    merged = get_selected(get_choices(state.prev_trace), complement(selection))
+    new_trace, weight = generate(branch_fn, args, merged)
+    retdiff = UnknownChange()
+    weight -= project(state.prev_trace, complement(selection))
+    weight += (project(new_trace, selection) - project(state.prev_trace, selection))
+    state.index = index
+    state.weight = weight
+    state.noise = project(new_trace, EmptySelection()) - project(state.prev_trace, EmptySelection())
+    state.score = get_score(new_trace)
+    state.trace = new_trace
+    state.retdiff = retdiff
+end
+
+function process!(gen_fn::Switch{C, N, K, T},
+                  index::Int,
+                  index_argdiff::NoChange,
+                  args::Tuple,
+                  kernel_argdiffs::Tuple,
+                  selection::Selection, 
+                  state::SwitchRegenerateState{T}) where {C, N, K, T}
+    new_trace, weight, retdiff = regenerate(getfield(state.prev_trace, :branch), args, kernel_argdiffs, selection)
+    state.index = index
+    state.weight = weight
+    state.noise = project(new_trace, EmptySelection()) - project(state.prev_trace, EmptySelection())
+    state.score = get_score(new_trace)
+    state.trace = new_trace
+    state.retdiff = retdiff
+end
+
+@inline process!(gen_fn::Switch{C, N, K, T}, index::C, index_argdiff::Diff, args::Tuple, kernel_argdiffs::Tuple, selection::Selection, state::SwitchRegenerateState{T}) where {C, N, K, T} = process!(gen_fn, getindex(gen_fn.cases, index), index_argdiff, args, kernel_argdiffs, selection, state)
+
+function regenerate(trace::SwitchTrace{T},
+                    args::Tuple, 
+                    argdiffs::Tuple,
+                    selection::Selection) where T
+    gen_fn = trace.gen_fn
+    index, index_argdiff = args[1], argdiffs[1]
+    state = SwitchRegenerateState{T}(0.0, 0.0, 0.0, trace)
+    process!(gen_fn, index, index_argdiff, args[2 : end], argdiffs[2 : end], selection, state)
+    return SwitchTrace(gen_fn, state.index, state.trace, get_retval(state.trace), args, state.score, state.noise), state.weight, state.retdiff
+end

src/modeling_library/switch/simulate.jl

@@ -0,0 +1,32 @@
+mutable struct SwitchSimulateState{T}
+    score::Float64
+    noise::Float64
+    index::Int
+    subtrace::Trace
+    retval::T
+    SwitchSimulateState{T}(score::Float64, noise::Float64) where T = new{T}(score, noise)
+end
+
+function process!(gen_fn::Switch{C, N, K, T},
+                  index::Int, 
+                  args::Tuple, 
+                  state::SwitchSimulateState{T}) where {C, N, K, T}
+    local retval::T
+    subtrace = simulate(getindex(gen_fn.branches, index), args)
+    state.index = index
+    state.noise += project(subtrace, EmptySelection())
+    state.subtrace = subtrace
+    state.score += get_score(subtrace)
+    state.retval = get_retval(subtrace)
+end
+
+@inline process!(gen_fn::Switch{C, N, K, T}, index::C, args::Tuple, state::SwitchSimulateState{T}) where {C, N, K, T} = process!(gen_fn, getindex(gen_fn.cases, index), args, state)
+
+function simulate(gen_fn::Switch{C, N, K, T},
+                  args::Tuple) where {C, N, K, T}
+
+    index = args[1]
+    state = SwitchSimulateState{T}(0.0, 0.0)
+    process!(gen_fn, index, args[2 : end], state)
+    return SwitchTrace{T}(gen_fn, state.index, state.subtrace, state.retval, args[2 : end], state.score, state.noise)
+end

src/modeling_library/switch/switch.jl

@@ -0,0 +1,56 @@
+struct Switch{C, N, K, T} <: GenerativeFunction{T, Trace}
+    branches::NTuple{N, GenerativeFunction{T}}
+    cases::Dict{C, Int}
+    function Switch(gen_fns::GenerativeFunction...)
+        @assert !isempty(gen_fns)
+        rettype = get_return_type(getindex(gen_fns, 1))
+        new{Int, length(gen_fns), typeof(gen_fns), rettype}(gen_fns, Dict{Int, Int}())
+    end
+    function Switch(d::Dict{C, Int}, gen_fns::GenerativeFunction...) where C
+        @assert !isempty(gen_fns)
+        rettype = get_return_type(getindex(gen_fns, 1))
+        new{C, length(gen_fns), typeof(gen_fns), rettype}(gen_fns, d)
+    end
+end
+export Switch
+
+has_argument_grads(switch_fn::Switch) = map(zip(map(has_argument_grads, switch_fn.branches)...)) do as
+    all(as)
+end
+accepts_output_grad(switch_fn::Switch) = all(accepts_output_grad, switch_fn.branches)
+
+function (gen_fn::Switch)(index::Int, args...)
+    (_, _, retval) = propose(gen_fn, (index, args...))
+    retval
+end
+
+function (gen_fn::Switch{C})(index::C, args...) where C
+    (_, _, retval) = propose(gen_fn, (gen_fn.cases[index], args...))
+    retval
+end
+
+include("assess.jl")
+include("propose.jl")
+include("simulate.jl")
+include("generate.jl")
+include("update.jl")
+include("regenerate.jl")
+include("backprop.jl")
+
+@doc(
+"""
+    gen_fn = Switch(gen_fns::GenerativeFunction...)
+
+Returns a new generative function that accepts an argument tuple of type `Tuple{Int, ...}` where the first index indicates which branch to call.
+
+    gen_fn = Switch(d::Dict{T, Int}, gen_fns::GenerativeFunction...) where T
+
+Returns a new generative function that accepts an argument tuple of type `Tuple{Int, ...}` or an argument tuple of type `Tuple{T, ...}` where the first index either indicates which branch to call, or indicates an index into `d` which maps to the selected branch. This form is meant for convenience - it allows the programmer to use `d` like if-else or case statements.
+
+`Switch` is designed to allow for the expression of patterns of if-else control flow. `gen_fns` must satisfy a few requirements:
+
+1. Each `gen_fn` in `gen_fns` must accept the same argument types.
+2. Each `gen_fn` in `gen_fns` must return the same return type.
+
+Otherwise, each `gen_fn` can come from different modeling languages, possess different traces, etc.
+""", Switch)