AliasMultinomial

AliasMultinomial.lua

@@ -0,0 +1,114 @@
+-- ref.: https://hips.seas.harvard.edu/blog/2013/03/03/the-alias-method-efficient-sampling-with-many-discrete-outcomes/
+local AM = torch.class("torch.AliasMultinomial") 
+
+function AM:__init(probs)
+   self.J, self.q = self:setup(probs)
+end
+
+function AM:setup(probs)
+   assert(probs:dim() == 1)
+   local K = probs:nElement()
+   local q = probs.new(K):zero()
+   local J = torch.LongTensor(K):zero()
+
+   -- Sort the data into the outcomes with probabilities
+   -- that are larger and smaller than 1/K.
+   local smaller, larger = {}, {}
+   local maxk, maxp = 0, -1
+   for kk = 1,K do
+      local prob = probs[kk]
+      q[kk] = K*prob
+      if q[kk] < 1 then
+         table.insert(smaller, kk)
+      else
+         table.insert(larger, kk)
+      end
+      if maxk > maxp then
+
+      end
+   end
+
+   -- Loop through and create little binary mixtures that
+   -- appropriately allocate the larger outcomes over the
+   -- overall uniform mixture.
+   while #smaller > 0 and #larger > 0 do
+      local small = table.remove(smaller)
+      local large = table.remove(larger)
+
+      J[small] = large
+      q[large] = q[large] - (1.0 - q[small])
+
+      if q[large] < 1.0 then
+         table.insert(smaller,large)
+      else
+         table.insert(larger,large)
+      end
+   end
+   assert(q:min() >= 0)
+   if q:max() > 1 then
+      q:div(q:max())
+   end
+   assert(q:max() <= 1)
+   if J:min() <= 0 then
+      -- sometimes an large index isn't added to J. 
+      -- fix it by making the probability 1 so that J isn't indexed.
+      local i = 0
+      J:apply(function(x)
+         i = i + 1
+         if x <= 0 then
+            q[i] = 1
+         end
+      end)
+   end
+   return J, q
+end
+
+function AM:draw()
+   J = self.J
+   q = self.q
+   local K  = J:nElement()
+
+   -- Draw from the overall uniform mixture.
+   local kk = math.random(1,K)
+
+   -- Draw from the binary mixture, either keeping the
+   -- small one, or choosing the associated larger one.
+   if math.random() < q[kk] then
+      return kk
+   else
+      return J[kk]
+   end
+end
+
+function AM:batchdraw(output)
+   assert(torch.type(output) == 'torch.LongTensor')
+   assert(output:nElement() > 0)
+   local J = self.J
+   local K  = J:nElement()
+
+   self._kk = self._kk or output.new()
+   self._kk:resizeAs(output):random(1,K)
+
+   self._q = self._q or self.q.new()
+   self._q:index(self.q, 1, self._kk:view(-1))
+
+   self._mask = self._b or torch.LongTensor()
+   self._mask:resize(self._q:size()):bernoulli(self._q)
+
+   self.__kk = self.__kk or output.new()
+   self.__kk:resize(self._kk:size()):copy(self._kk)
+   self.__kk:cmul(self._mask)
+
+   -- if mask == 0 then output[i] = J[kk[i]] else output[i] = 0
+
+   self._mask:add(-1):mul(-1) -- (1,0) - > (0,1)
+   output:view(-1):index(J, 1, self._kk:view(-1))
+   output:cmul(self._mask)
+
+   -- elseif mask == 1 then output[i] = kk[i]
+
+   output:add(self.__kk)
+
+   return output
+end
+

init.lua

@@ -17,6 +17,7 @@ torch.include('torchx', 'indexdir.lua')
 torch.include('torchx', 'dkjson.lua')
 torch.include('torchx', 'recursivetensor.lua')
 torch.include('torchx', 'Queue.lua')
+torch.include('torchx', 'AliasMultinomial.lua')
 
 torch.include('torchx', 'test.lua')
 

test/test.lua

@@ -104,6 +104,33 @@ function torchxtest.concat()
    mytester:assertTensorEq(res,res2,0.00001)
 end
 
+function torchxtest.AliasMultinomial()
+   local probs = torch.Tensor(10):uniform(0,1)
+   probs:div(probs:sum())
+
+   local a = torch.Timer()
+   local am = torch.AliasMultinomial(probs)
+   print("setup in "..a:time().real.." seconds")
+
+   a:reset()
+   am:draw()
+   print("draw in "..a:time().real.." seconds")
+
+   local output = torch.LongTensor(1000, 1000)
+   a:reset()
+   am:batchdraw(output)
+   print("batchdraw in "..a:time().real.." seconds")
+
+   local counts = torch.Tensor(10):zero()
+   output:apply(function(x)
+      counts[x] = counts[x] + 1
+   end)
+
+   counts:div(counts:sum())
+
+   mytester:assertTensorEq(probs, counts, 0.001)
+end
+
 function torchx.test(tests)
    local oldtype = torch.getdefaulttensortype()
    torch.setdefaulttensortype('torch.FloatTensor')