wip

Author: chentoast

src/inference/hmc_common.jl

@@ -9,3 +9,44 @@ function assess_momenta(momenta)
   end
   logprob
 end
+
+function add_choicemaps(a::ChoiceMap, b::ChoiceMap)
+  out = choicemap()
+
+  for (name, val) in get_values_shallow(a)
+    out[name] = val + b[name]
+  end
+
+  for (name, submap) in get_submaps_shallow(a)
+    out.internal_nodes[name] = add_choicemaps(submap, get_submap(b, name))
+  end
+
+  return out
+end
+
+function scale_choicemap(a::ChoiceMap, scale)
+  out = choicemap()
+
+  for (name, val) in get_values_shallow(a)
+    out[name] = val * scale
+  end
+
+  for (name, submap) in get_submaps_shallow(a)
+    out.internal_nodes[name] = scale_choicemap(submap, scale)
+  end
+
+  return out
+end
+
+function assess_momenta_trie(momenta_trie)
+  logprob = 0.
+  for (_, val) in get_values_shallow(momenta_trie)
+    logprob += logpdf(normal, val, 0, 1)
+  end
+
+  for (_, submap) in get_submaps_shallow(momenta_trie)
+    logprob += assess_momenta_trie(submap)
+  end
+
+  return logprob
+end

src/inference/nuts.jl

@@ -15,6 +15,7 @@ end
 struct SamplerStats
   depth
   n
+  diverging
   accept
 end
 
@@ -23,34 +24,30 @@ function u_turn(values_left, values_right, momenta_left, momenta_right)
   (dot(values_right - values_left, momenta_left) >= 0)
 end
 
-function leapfrog(values, momenta, eps, integrator_state)
-  values_trie, selection, retval_grad, trace = integrator_state
+function leapfrog(values_trie, momenta_trie, eps, integrator_state)
+  selection, retval_grad, trace = integrator_state
 
-  values_trie = from_array(values_trie, values)
   (trace, _, _) = update(trace, values_trie)
   (_, _, gradient_trie) = choice_gradients(trace, selection, retval_grad)
-  gradient = to_array(gradient_trie, Float64)
 
   # half step on momenta
-  momenta += (eps / 2) * gradient
+  momenta_trie = add_choicemaps(momenta_trie, scale_choicemap(gradient_trie, eps / 2))
 
   # full step on positions
-  values += eps * momenta
+  values_trie = add_choicemaps(values_trie, scale_choicemap(momenta_trie, eps))
 
   # get new gradient
-  values_trie = from_array(values_trie, values)
   (trace, _, _) = update(trace, values_trie)
   (_, _, gradient_trie) = choice_gradients(trace, selection, retval_grad)
-  gradient = to_array(gradient_trie, Float64)
 
   # half step on momenta
-  momenta += (eps / 2) * gradient
-  return values, momenta, get_score(trace)
+  momenta_trie = add_choicemaps(momenta_trie, scale_choicemap(gradient_trie, eps / 2))
+  return values_trie, momenta_trie, get_score(trace)
 end
 
 function build_root(val, momenta, eps, direction, weight_init, integrator_state)
   val, momenta, lp = leapfrog(val, momenta, direction * eps, integrator_state)
-  weight = lp + assess_momenta(momenta)
+  weight = lp + assess_momenta_trie(momenta)
 
   diverging = weight - weight_init > 1000
 
@@ -67,9 +64,16 @@ function merge_trees(tree_left, tree_right)
 
   weight = logsumexp(tree_left.weight, tree_right.weight)
   n = tree_left.n + tree_right.n
-  stop = tree_left.stop || tree_right.stop || u_turn(
-    tree_left.val_left, tree_right.val_right, tree_left.momenta_left, tree_right.momenta_right
-  )
+  
+  if u_turn(to_array(tree_left.val_left, Float64),
+    to_array(tree_right.val_right, Float64),
+    to_array(tree_left.momenta_left, Float64),
+    to_array(tree_right.momenta_right, Float64))
+  end
+  stop = tree_left.stop || tree_right.stop || u_turn(to_array(tree_left.val_left, Float64),
+                                                     to_array(tree_right.val_right, Float64),
+                                                     to_array(tree_left.momenta_left, Float64),
+                                                     to_array(tree_right.momenta_right, Float64))
   diverging = tree_left.diverging || tree_right.diverging
 
   return Tree(tree_left.val_left, tree_left.momenta_left, tree_right.val_right,
@@ -90,14 +94,10 @@ function build_tree(val, momenta, depth, eps, direction, weight_init, integrator
   if direction == 1
     other_tree = build_tree(tree.val_right, tree.momenta_right, depth - 1, eps, direction,
                             weight_init, integrator_state)
+    return merge_trees(tree, other_tree)
   else
     other_tree = build_tree(tree.val_left, tree.momenta_left, depth - 1, eps, direction,
                             weight_init, integrator_state)
-  end
-
-  if direction == 1
-    return merge_trees(tree, other_tree)
-  else
     return merge_trees(other_tree, tree)
   end
 end
@@ -131,16 +131,16 @@ function nuts(
 
     # values needed for a leapfrog step
     (_, values_trie, _) = choice_gradients(trace, selection, retval_grad)
-    values = to_array(values_trie, Float64)
 
-    momenta = sample_momenta(length(values))
+    momenta = sample_momenta(length(to_array(values_trie, Float64)))
+    momenta_trie = from_array(values_trie, momenta)
     prev_momenta_score = assess_momenta(momenta)
 
     weight_init = prev_model_score + prev_momenta_score
 
-    integrator_state = (values_trie, selection, retval_grad, trace)
+    integrator_state = (selection, retval_grad, trace)
 
-    tree = Tree(values, momenta, values, momenta, values, 1, -Inf, false, false)
+    tree = Tree(values_trie, momenta_trie, values_trie, momenta_trie, values_trie, 1, -Inf, false, false)
 
     direction = 0
     depth = 0
@@ -165,25 +165,25 @@ function nuts(
       depth += 1
     end
 
-    (new_trace, _, _) = update(trace, from_array(values_trie, tree.val_sample))
+    (new_trace, _, _) = update(trace, tree.val_sample)
     check && check_observations(get_choices(new_trace), observations)
 
     # assess new model score (negative potential energy)
     new_model_score = get_score(new_trace)
 
     # assess new momenta score (negative kinetic energy)
     if direction == 1
-      new_momenta_score = assess_momenta(-tree.momenta_right)
+      new_momenta_score = assess_momenta_trie(tree.momenta_right)
     else
-      new_momenta_score = assess_momenta(-tree.momenta_left)
+      new_momenta_score = assess_momenta_trie(tree.momenta_left)
     end
 
     # accept or reject
     alpha = new_model_score + new_momenta_score - weight_init
     if log(rand()) < alpha
-      return (new_trace, SamplerStats(depth, tree.n, true))
+      return (new_trace, SamplerStats(depth, tree.n, tree.diverging, true))
     else
-      return (trace, SamplerStats(depth, tree.n, false))
+      return (trace, SamplerStats(depth, tree.n, tree.diverging, false))
     end
 end
 

test/inference/nuts.jl

@@ -0,0 +1,20 @@
+@testset "nuts" begin
+
+  # smoke test a function without retval gradient
+  @gen function foo()
+      x = @trace(normal(0, 1), :x)
+      return x
+  end
+
+  (trace, _) = generate(foo, ())
+  (new_trace, accepted) = nuts(trace, select(:x))
+
+  # smoke test a function with retval gradient
+  @gen (grad) function foo()
+      x = @trace(normal(0, 1), :x)
+      return x
+  end
+
+  (trace, _) = generate(foo, ())
+  (new_trace, accepted) = nuts(trace, select(:x))
+end