Skip to content

Commit

Permalink
Merge pull request #373 from JuliaRobotics/maint/20Q1/refactCopy300
Browse files Browse the repository at this point in the history
Refactor graph copy and subgraph functions
  • Loading branch information
Affie authored Apr 13, 2020
2 parents 86c7ae6 + c85774a commit d708217
Show file tree
Hide file tree
Showing 14 changed files with 530 additions and 409 deletions.
68 changes: 0 additions & 68 deletions attic/sandbox.jl

This file was deleted.

2 changes: 1 addition & 1 deletion src/DFGPlots/DFGPlots.jl
Original file line number Diff line number Diff line change
Expand Up @@ -88,7 +88,7 @@ More information at [GraphPlot.jl](https://github.com/JuliaGraphs/GraphPlot.jl)
function dfgplot(dfg::AbstractDFG, p::DFGPlotProps = DFGPlotProps())
# TODO implement convert functions
ldfg = LightDFG{NoSolverParams}()
DistributedFactorGraphs._copyIntoGraph!(dfg, ldfg, union(listVariables(dfg), listFactors(dfg)), true, copyGraphMetadata=false)
copyGraph!(ldfg, dfg, union(listVariables(dfg), listFactors(dfg)), copyGraphMetadata=false)
dfgplot(ldfg, p)
end

Expand Down
197 changes: 196 additions & 1 deletion src/Deprecated.jl
Original file line number Diff line number Diff line change
Expand Up @@ -95,4 +95,199 @@ Base.setproperty!(x::DFGFactor,f::Symbol, val) = begin
end


#
#NOTE deprecate in favor of constructors because its not lossless: https://docs.julialang.org/en/v1/manual/conversion-and-promotion/#Conversion-vs.-Construction-1
function Base.convert(::Type{DFGVariableSummary}, v::DFGVariable)
Base.depwarn("convert to type DFGVariableSummary is deprecated use the constructor", :convert)
return DFGVariableSummary(v)
end

function Base.convert(::Type{SkeletonDFGVariable}, v::VariableDataLevel1)
Base.depwarn("convert to type SkeletonDFGVariable is deprecated use the constructor", :convert)
return SkeletonDFGVariable(v)
end

function Base.convert(::Type{DFGFactorSummary}, f::DFGFactor)
Base.depwarn("convert to type DFGFactorSummary is deprecated use the constructor", :convert)
return DFGFactorSummary(f)
end

function Base.convert(::Type{SkeletonDFGFactor}, f::FactorDataLevel1)
Base.depwarn("convert to type SkeletonDFGFactor is deprecated use the constructor", :convert)
return SkeletonDFGFactor(f)
end

##==============================================================================
## WIP on consolidated subgraph functions, aim to remove in 0.8
##==============================================================================
# Deprecate in favor of buildSubgraph, mergeGraph
export getSubgraph, getSubgraphAroundNode
export buildSubgraphFromLabels!

"""
$(SIGNATURES)
Common function for copying nodes from one graph into another graph.
This is overridden in specialized implementations for performance.
NOTE: copyGraphMetadata not supported yet.
"""
function _copyIntoGraph!(sourceDFG::G, destDFG::H, variableFactorLabels::Vector{Symbol}, includeOrphanFactors::Bool=false; copyGraphMetadata::Bool=false)::Nothing where {G <: AbstractDFG, H <: AbstractDFG}
# Split into variables and factors
Base.depwarn("_copyIntoGraph! is deprecated use copyGraph/deepcopyGraph[!]", :_copyIntoGraph!)

includeOrphanFactors && (@error "Adding orphaned factors is not supported")

sourceVariables = map(vId->getVariable(sourceDFG, vId), intersect(listVariables(sourceDFG), variableFactorLabels))
sourceFactors = map(fId->getFactor(sourceDFG, fId), intersect(listFactors(sourceDFG), variableFactorLabels))
if length(sourceVariables) + length(sourceFactors) != length(variableFactorLabels)
rem = symdiff(map(v->v.label, sourceVariables), variableFactorLabels)
rem = symdiff(map(f->f.label, sourceFactors), variableFactorLabels)
error("Cannot copy because cannot find the following nodes in the source graph: $rem")
end

# Now we have to add all variables first,
for variable in sourceVariables
if !exists(destDFG, variable)
addVariable!(destDFG, deepcopy(variable))
end
end
# And then all factors to the destDFG.
for factor in sourceFactors
# Get the original factor variables (we need them to create it)
sourceFactorVariableIds = getNeighbors(sourceDFG, factor)
# Find the labels and associated variables in our new subgraph
factVariableIds = Symbol[]
for variable in sourceFactorVariableIds
if exists(destDFG, variable)
push!(factVariableIds, variable)
end
end
# Only if we have all of them should we add it (otherwise strange things may happen on evaluation)
if includeOrphanFactors || length(factVariableIds) == length(sourceFactorVariableIds)
if !exists(destDFG, factor)
addFactor!(destDFG, deepcopy(factor))
end
end
end

if copyGraphMetadata
setUserData(destDFG, getUserData(sourceDFG))
setRobotData(destDFG, getRobotData(sourceDFG))
setSessionData(destDFG, getSessionData(sourceDFG))
end
return nothing
end

"""
$(SIGNATURES)
Retrieve a deep subgraph copy around a given variable or factor.
Optionally provide a distance to specify the number of edges should be followed.
Optionally provide an existing subgraph addToDFG, the extracted nodes will be copied into this graph. By default a new subgraph will be created.
Note: By default orphaned factors (where the subgraph does not contain all the related variables) are not returned. Set includeOrphanFactors to return the orphans irrespective of whether the subgraph contains all the variables.
Note: Always returns the node at the center, but filters around it if solvable is set.
"""
function getSubgraphAroundNode(dfg::AbstractDFG, node::DFGNode, distance::Int=1, includeOrphanFactors::Bool=false, addToDFG::AbstractDFG=_getDuplicatedEmptyDFG(dfg); solvable::Int=0)::AbstractDFG

Base.depwarn("getSubgraphAroundNode is deprecated use buildSubgraph", :getSubgraphAroundNode)

if !exists(dfg, node.label)
error("Variable/factor with label '$(node.label)' does not exist in the factor graph")
end

neighbors = getNeighborhood(dfg, node.label, distance)

# for some reason: always returns the node at the center with || (nlbl == node.label)
solvable != 0 && filter!(nlbl -> (getSolvable(dfg, nlbl) >= solvable) || (nlbl == node.label), neighbors)

# Copy the section of graph we want
_copyIntoGraph!(dfg, addToDFG, neighbors, includeOrphanFactors)
return addToDFG
end

"""
$(SIGNATURES)
Get a deep subgraph copy from the DFG given a list of variables and factors.
Optionally provide an existing subgraph addToDFG, the extracted nodes will be copied into this graph. By default a new subgraph will be created.
Note: By default orphaned factors (where the subgraph does not contain all the related variables) are not returned. Set includeOrphanFactors to return the orphans irrespective of whether the subgraph contains all the variables.
"""
function getSubgraph(dfg::G,
variableFactorLabels::Vector{Symbol},
includeOrphanFactors::Bool=false,
addToDFG::H=_getDuplicatedEmptyDFG(dfg))::H where {G <: AbstractDFG, H <: AbstractDFG}

Base.depwarn("getSubgraph is deprecated use buildSubgraph", :getSubgraph)

for label in variableFactorLabels
if !exists(dfg, label)
error("Variable/factor with label '$(label)' does not exist in the factor graph")
end
end

_copyIntoGraph!(dfg, addToDFG, variableFactorLabels, includeOrphanFactors)
return addToDFG
end


# TODO needsahome: home should be in IIF, calling just deepcopyGraph, or copyGraph
# Into, Labels, Subgraph are all implied from the parameters.
# can alies names but like Sam suggested only on copy is needed.


"""
$SIGNATURES
Construct a new factor graph object as a subgraph of `dfg <: AbstractDFG` based on the
variable labels `syms::Vector{Symbols}`.
SamC: Can we not just use _copyIntoGraph! for this? Looks like a small refactor to make it work.
Will paste in as-is for now and we can figure it out as we go.
DF: Absolutely agree that subgraph functions should use `DFG._copyIntoGraph!` as a single dependency in the code base. There have been a repeated new rewrites of IIF.buildSubGraphFromLabels (basic wrapper is fine) but nominal should be to NOT duplicate DFG functionality in IIF -- rather use/improve the existing features in DFG. FYI, I have repeatedly refactored this function over and over to use DFG properly but somehow this (add/delete/Variable/Factor) version keeps coming back without using `_copyIntoGraph`!!??? See latest effort commented out below `buildSubgraphFromLabels!_SPECIAL`...
Notes
- Slighly messy internals, but gets the job done -- some room for performance improvement.
- Defaults to GraphDFG, but likely to change to LightDFG in future.
- since DFG v0.6 LightDFG is the default.
DevNotes
- TODO: still needs to be consolidated with `DFG._copyIntoGraph`
Related
listVariables, _copyIntoGraph!
"""
function buildSubgraphFromLabels!(dfg::G,
syms::Vector{Symbol};
subfg::AbstractDFG=(G <: InMemoryDFGTypes ? G : GraphsDFG)(params=getSolverParams(dfg)),
solvable::Int=0,
allowedFactors::Union{Nothing, Vector{Symbol}}=nothing )::AbstractDFG where G <: AbstractDFG
#
Base.depwarn("buildSubgraphFromLabels! is deprecated use copyGraph, buildSubgraph or buildCliqueSubgraph!(IIF)", :buildSubgraphFromLabels!)
# add a little too many variables (since we need the factors)
for sym in syms
if solvable <= getSolvable(dfg, sym)
getSubgraphAroundNode(dfg, getVariable(dfg, sym), 2, false, subfg, solvable=solvable)
end
end

# remove excessive variables that were copied by neighbors distance 2
currVars = listVariables(subfg)
toDelVars = setdiff(currVars, syms)
for dv in toDelVars
# delete any neighboring factors first
for fc in lsf(subfg, dv)
deleteFactor!(subfg, fc)
end

# and the variable itself
deleteVariable!(subfg, dv)
end

# delete any factors not in the allowed list
if allowedFactors != nothing
delFcts = setdiff(lsf(subfg), allowedFactors)
for dfct in delFcts
deleteFactor!(subfg, dfct)
end
end

# orphaned variables are allowed, but not orphaned factors

return subfg
end
9 changes: 3 additions & 6 deletions src/DistributedFactorGraphs.jl
Original file line number Diff line number Diff line change
Expand Up @@ -161,8 +161,8 @@ export packVariableNodeData, unpackVariableNodeData

export getSolvedCount, isSolved, setSolvedCount!, isInitialized

export getNeighborhood, getSubgraph, getSubgraphAroundNode, getNeighbors, _getDuplicatedEmptyDFG

export getNeighborhood, getNeighbors, _getDuplicatedEmptyDFG
export copyGraph!, deepcopyGraph, deepcopyGraph!, buildSubgraph, mergeGraph!
# Big Data
##------------------------------------------------------------------------------
export addBigDataEntry!, getBigDataEntry, updateBigDataEntry!, deleteBigDataEntry!, getBigDataEntries, getBigDataKeys
Expand Down Expand Up @@ -224,13 +224,10 @@ export
compareFactorGraphs


## Deprecated.jl should be listed there
## Deprecated exports should be listed in Deprecated.jl if possible, otherwise here


## needsahome.jl

export buildSubgraphFromLabels!

import Base: print
export printFactor, printVariable, print

Expand Down
38 changes: 1 addition & 37 deletions src/GraphsDFG/services/GraphsDFG.jl
Original file line number Diff line number Diff line change
Expand Up @@ -208,7 +208,7 @@ function getFactors(dfg::GraphsDFG, regexFilter::Union{Nothing, Regex}=nothing;
if regexFilter != nothing
factors = filter(f -> occursin(regexFilter, String(f.label)), factors)
end

if length(tags) > 0
mask = map(v -> length(intersect(v.tags, tags)) > 0, factors )
return factors[mask]
Expand Down Expand Up @@ -255,42 +255,6 @@ function getNeighbors(dfg::GraphsDFG, label::Symbol; solvable::Int=0)::Vector{Sy
return map(n -> n.dfgNode.label, neighbors)
end

#NOTE Replaced by abstract function in services/AbstractDFG.jl
# """
# $(SIGNATURES)
# Retrieve a deep subgraph copy around a given variable or factor.
# Optionally provide a distance to specify the number of edges should be followed.
# Optionally provide an existing subgraph addToDFG, the extracted nodes will be copied into this graph. By default a new subgraph will be created.
# Note: By default orphaned factors (where the subgraph does not contain all the related variables) are not returned. Set includeOrphanFactors to return the orphans irrespective of whether the subgraph contains all the variables.
# """
# function getSubgraphAroundNode(dfg::GraphsDFG{P}, node::T, distance::Int64=1, includeOrphanFactors::Bool=false, addToDFG::GraphsDFG=GraphsDFG{P}(); solvable::Int=0)::GraphsDFG where {P <: AbstractParams, T <: DFGNode}
# if !haskey(dfg.labelDict, node.label)
# error("Variable/factor with label '$(node.label)' does not exist in the factor graph")
# end
#
# # Build a list of all unique neighbors inside 'distance'
# neighborList = Dict{Symbol, Any}()
# push!(neighborList, node.label => dfg.g.vertices[dfg.labelDict[node.label]])
# curList = Dict{Symbol, Any}(node.label => dfg.g.vertices[dfg.labelDict[node.label]])
# for dist in 1:distance
# newNeighbors = Dict{Symbol, Any}()
# for (key, node) in curList
# neighbors = in_neighbors(node, dfg.g) #Don't use out_neighbors! It enforces directiveness even if we don't want it
# for neighbor in neighbors
# if !haskey(neighborList, neighbor.dfgNode.label) && (isSolvable(neighbor.dfgNode) >= solvable)
# push!(neighborList, neighbor.dfgNode.label => neighbor)
# push!(newNeighbors, neighbor.dfgNode.label => neighbor)
# end
# end
# end
# curList = newNeighbors
# end
#
# # Copy the section of graph we want
# _copyIntoGraph!(dfg, addToDFG, collect(keys(neighborList)), includeOrphanFactors)
# return addToDFG
# end

"""
$(SIGNATURES)
Produces a dot-format of the graph for visualization.
Expand Down
8 changes: 5 additions & 3 deletions src/LightDFG/LightDFG.jl
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,9 @@ module LightDFGs
using LightGraphs
using DocStringExtensions

import ...DistributedFactorGraphs: AbstractDFG, DFGNode, AbstractDFGVariable, AbstractDFGFactor, DFGSummary, AbstractParams, NoSolverParams, DFGVariable, DFGFactor
using ...DistributedFactorGraphs

# import ...DistributedFactorGraphs: AbstractDFG, DFGNode, AbstractDFGVariable, AbstractDFGFactor, DFGSummary, AbstractParams, NoSolverParams, DFGVariable, DFGFactor

# import DFG functions to extend
import ...DistributedFactorGraphs: setSolverParams!,
Expand Down Expand Up @@ -38,8 +40,8 @@ import ...DistributedFactorGraphs: setSolverParams!,
isFullyConnected,
hasOrphans,
getNeighbors,
getSubgraphAroundNode,
getSubgraph,
buildSubgraph,
copyGraph!,
getBiadjacencyMatrix,
_getDuplicatedEmptyDFG,
toDot,
Expand Down
11 changes: 11 additions & 0 deletions src/LightDFG/entities/LightDFG.jl
Original file line number Diff line number Diff line change
Expand Up @@ -66,3 +66,14 @@ LightDFG(description::String,
sessionData::Dict{Symbol, String},
solverParams::AbstractParams) =
LightDFG(FactorGraph{Int,DFGVariable,DFGFactor}(), description, userId, robotId, sessionId, userData, robotData, sessionData, Symbol[], solverParams)


LightDFG{T,V,F}(description::String,
userId::String,
robotId::String,
sessionId::String,
userData::Dict{Symbol, String},
robotData::Dict{Symbol, String},
sessionData::Dict{Symbol, String},
solverParams::T) where {T <: AbstractParams, V <:AbstractDFGVariable, F<:AbstractDFGFactor} =
LightDFG(FactorGraph{Int,V,F}(), description, userId, robotId, sessionId, userData, robotData, sessionData, Symbol[], solverParams)
Loading

0 comments on commit d708217

Please sign in to comment.