-#!/usr/bin/env luajit
require 'torch'
require 'nn'
-require 'image'
-require 'optim'
-----------------------------------------------------------------------
+local DAG, parent = torch.class('nn.DAG', 'nn.Container')
-local Graph, parent = torch.class('nn.Graph', 'nn.Container')
-
-function Graph:__init()
+function DAG:__init()
parent.__init(self)
- self.pred = {}
- self.succ = {}
+ -- Nodes are indexed by the module they encompass
+ self.node = { }
end
-function Graph:addEdge(a, b)
- local pred, succ = self.pred, self.succ
- if not pred[a] and not succ[a] then
- self:add(a)
- end
- if not pred[b] and not succ[b] then
- self:add(b)
+function DAG:createNode(n)
+ if not self.node[n] then
+ self:add(n) -- Add it to the object as a Container
+ self.node[n] = {}
+ self.node[n].succ = {}
+ self.node[n].pred = {}
end
- pred[b] = pred[b] or {}
- pred[b][#pred[b] + 1] = a
- succ[a] = succ[a] or {}
- succ[a][#succ[a] + 1] = b
end
-function Graph:setInput(i)
- if torch.type(i) == 'table' then
- self.inputModules = i
- for _, m in ipairs(i) do
- if not self.pred[m] and not self.succ[m] then
- self:add(m)
- end
+function DAG:addEdge(a, b)
+ self.sorted = nil
+ self:createNode(a)
+ self:createNode(b)
+ table.insert(self.node[b].pred, a)
+ table.insert(self.node[a].succ, b)
+end
+
+-- Apply f on t recursively; use the corresponding a1 and a2 elements
+-- (i.e. same keys) as second and third parameters to f when
+-- available; return the results from f, organized in a similarly
+-- nested table.
+function DAG:nestApply(f, t, a1, a2)
+ if torch.type(t) == 'table' then
+ local result = {}
+ for k, s in pairs(t) do
+ result[k] = self:nestApply(f, s, a1 and a1[k], a2 and a2[k])
end
+ return result
else
- self:setInput({ i })
+ return f(t, a1, a2)
end
end
-function Graph:setOutput(o)
- if torch.type(o) == 'table' then
- self.outputModules = o
- for _, m in ipairs(o) do
- if not self.pred[m] and not self.succ[m] then
- self:add(m)
+function DAG:setInput(i)
+ self.sorted = nil
+ self.inputModules = i
+ self:nestApply(
+ function(m)
+ if #self.node[m].succ == 0 then
+ error('Input modules must have outgoing edges.')
end
- end
- else
- self:setOutput({ o })
- end
+ if #self.node[m].pred > 0 then
+ error('Input modules cannog have incoming edges.')
+ end
+ end,
+ self.inputModules
+ )
end
-function Graph:order()
- local distance = {}
+function DAG:setOutput(o)
+ self.sorted = nil
+ self.outputModules = o
+ self:nestApply(
+ function(m)
+ if #self.node[m].pred == 0 then
+ error('Output module must have incoming edges.')
+ end
+ if #self.node[m].succ > 0 then
+ error('Output module cannot have outgoing edges.')
+ end
+ end,
+ self.outputModules
+ )
+end
- for _, a in pairs(self.inputModules) do
- distance[a] = 1
+function DAG:putInOrder()
+ if self.sorted then
+ return
end
+ -- First, we sort the nodes according to the DAG order
+
+ local distance = {}
+
+ self:nestApply(function(m) distance[m] = 1 end, self.inputModules)
+
local nc
repeat
nc = 0
- for i, isucc in pairs(self.succ) do
- for _, j in pairs(isucc) do
+ for i, node in pairs(self.node) do
+ for _, j in pairs(node.succ) do
if distance[i] and (not distance[j] or distance[j] < distance[i] + 1) then
distance[j] = distance[i] + 1
nc = nc + 1
until nc == 0
self.sorted = { }
- for i, d in pairs(distance) do
- table.insert(self.sorted, { d, i })
+ for n, d in pairs(distance) do
+ table.insert(self.sorted, { distance = d, node = n })
end
- table.sort(self.sorted, function(a, b) return a[1] < b[1] end)
- for i, a in ipairs(self.sorted) do self.sorted[i] = a[2] end
+ table.sort(self.sorted, function(a, b) return a.distance < b.distance end)
+
+ for i, a in ipairs(self.sorted) do self.sorted[i] = a.node end
end
-function Graph:print()
+function DAG:print()
+ self:putInOrder()
+
for i, d in ipairs(self.sorted) do
print('#' .. i .. ' -> ' .. torch.type(d))
end
end
-function Graph:updateOutput(input)
- if #self.inputModules == 1 then
- self.inputModules[1]:updateOutput(input)
- else
- for i, d in ipairs(self.inputModules) do
- d:updateOutput(input[i])
- end
- end
+function DAG:updateOutput(input)
+ self:putInOrder()
- for _, d in ipairs(self.sorted) do
- if self.pred[d] then
- if #self.pred[d] == 1 then
- d:updateOutput(self.pred[d][1].output)
- elseif #self.pred[d] > 1 then
- local c = {}
- for k = 1, #self.pred[d] do
- c[k] = self.pred[d][k].output
+ self:nestApply(function(m, i) m:updateOutput(i) end, self.inputModules, input)
+
+ for _, m in ipairs(self.sorted) do
+ if #self.node[m].pred > 0 then
+ local i
+ if #self.node[m].pred == 1 then
+ i = self.node[m].pred[1].output
+ elseif #self.node[m].pred > 1 then
+ i = {}
+ for k = 1, #self.node[m].pred do
+ i[k] = self.node[m].pred[k].output
end
- d:updateOutput(c)
end
+ self.node[m].input = i
+ m:updateOutput(i)
end
end
- if #self.outputModules == 1 then
- self.output = self.outputModules[1].output
- else
- self.output = { }
- for i, d in ipairs(self.outputModules) do
- self.output[i] = d.output
- end
- end
+ self.output = self:nestApply(function(m) return m.output end, self.outputModules)
return self.output
end
-----------------------------------------------------------------------
-
-a = nn.Linear(10, 10)
-b = nn.ReLU()
-c = nn.Linear(10, 3)
-d = nn.Linear(10, 3)
-e = nn.CMulTable()
-f = nn.Linear(3, 2)
-
---[[
+function DAG:updateGradInput(input, gradOutput)
+ self:putInOrder()
- a -----> b ---> c ----> e ---
- \ /
- \--> d ---/
- \
- \---> f ---
-]]--
+ self:nestApply(
+ function(m, go) m:updateGradInput(self.node[m].input, go) end,
+ self.outputModules, gradOutput
+ )
-g = Graph:new()
-
-g:setInput(a)
-g:setOutput({ e, f })
-g:addEdge(c, e)
-g:addEdge(a, b)
-g:addEdge(d, e)
-g:addEdge(b, c)
-g:addEdge(b, d)
-g:addEdge(d, f)
+ for _, node in pairs(self.node) do
+ node.gradInputSucc = {}
+ end
-g:order()
+ for k = #self.sorted, 1, -1 do
+ local m = self.sorted[k]
+ local node = self.node[m]
+ local pred, succ, gradInputSucc = node.pred, node.succ, node.gradInputSucc
+
+ -- We update m:gradInput
+ if #gradInputSucc == 1 then
+ m:updateGradInput(node.input, gradInputSucc[1])
+ elseif #gradInputSucc > 1 then
+ local sum
+ for k = 1, #succ do
+ if sum then
+ sum:add(succ[k].gradInput)
+ else
+ sum = succ[k].gradInput
+ end
+ end
+ m:updateGradInput(node.input, sum)
+ end
-g:print(graph)
+ -- We fill the gradInputSucc of our predecessors
+ if #pred == 1 then
+ table.insert(self.node[pred[1]].gradInputSucc, node.gradInput)
+ elseif #pred > 1 then
+ for n = 1, #pred do
+ table.insert(self.node[node.pred[n]].gradInputSucc, m.gradInput[n])
+ end
+ end
+ end
-input = torch.Tensor(3, 10):uniform()
+ self.gradInput = self:nestApply(function(m) return m.gradInput end, self.inputModules)
-output = g:updateOutput(input)
+ return self.gradInput
+end
-print(output[1])
-print(output[2])
+return DAG
projects / dagnn.git / blobdiff