Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Split iteration in forwardpass and backwardpass #13

Merged
merged 1 commit into from
Aug 25, 2017
Merged

Split iteration in forwardpass and backwardpass #13

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
66 changes: 39 additions & 27 deletions src/sddp.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,16 +41,7 @@ function applycuts(pathsd, g)
end
end

"""
$(SIGNATURES)

Runs one iteration of the SDDP algorithm on the lattice given by `g`.
A total of `Ktot` paths will be explored up to `num_stages` stages.
The paths will be selected according to `pathsampler` and equivalent paths might be merged if their difference is smaller than `ztol` and `mergepaths` is true.
The parameter `ztol` is also used to check whether a new cut is useful.
When a scenario is infeasible and `stopatinf` is true then no other scenario with the same ancestor is run. Note that since the order in which the different scenarios is run is not deterministic, this might introduce nondeterminism even if the sampling is deterministic.
"""
function iteration(g::AbstractSDDPGraph{S}, Ktot::Int, num_stages, verbose, pathsampler; ztol=1e-6, stopatinf=false, mergepaths=true, forwardcuts=false, backwardcuts=true) where S
function forwardpass!(g::AbstractSDDPGraph, Ktot::Int, num_stages, verbose, pathsampler; ztol=1e-6, stopatinf=false, mergepaths=true, forwardcuts=false)
stats = SDDPStats()

master, initialnode = getmaster(g)
Expand Down Expand Up @@ -102,28 +93,49 @@ function iteration(g::AbstractSDDPGraph{S}, Ktot::Int, num_stages, verbose, path
z_UB, σ = meanstdpaths(endedpaths, Ktot)
end

if backwardcuts # We could ask a node if it has new fcuts/ocuts before resolving, that way the last stage won't be a particular case anymore
for t in num_stages:-1:2
# The last stages do not need to be resolved since it does not have any new cut
if t != num_stages
# Make jobs
jobsd = childjobs(g, pathsd[t-1], pathsampler, t, num_stages)
# Solve Jobs (parallelism possible here)
infeasibility_detected |= solvejobs!(jobsd, stats, stopatinf)
end

gencuts(pathsd[t-1], g, stats, ztol)
# The cut is added after so that they are added by group and duplicate detection in CutPruners works better
applycuts(pathsd[t-1], g)
end
end

# update stats
stats.upperbound = z_UB
stats.σ_UB = σ
stats.npaths = Ktot
stats.lowerbound = mastersol.objval

pathsd, mastersol, stats
end

function backwardpass!(g::AbstractSDDPGraph, num_stages, pathsd, pathsampler, stats; ztol=1e-6, stopatinf=false)
# We could ask a node if it has new fcuts/ocuts before resolving, that way the last stage won't be a particular case anymore
for t in num_stages:-1:2
# The last stages do not need to be resolved since it does not have any new cut
if t != num_stages
# Make jobs
jobsd = childjobs(g, pathsd[t-1], pathsampler, t, num_stages) # FIXME shouldn't need pathsampler here
# Solve Jobs (parallelism possible here)
solvejobs!(jobsd, stats, stopatinf)
end

gencuts(pathsd[t-1], g, stats, ztol)
# The cut is added after so that they are added by group and duplicate detection in CutPruners works better
applycuts(pathsd[t-1], g)
end
end

"""
$(SIGNATURES)

Runs one iteration of the SDDP algorithm on the lattice given by `g`.
A total of `Ktot` paths will be explored up to `num_stages` stages.
The paths will be selected according to `pathsampler` and equivalent paths might be merged if their difference is smaller than `ztol` and `mergepaths` is true.
The parameter `ztol` is also used to check whether a new cut is useful.
When a scenario is infeasible and `stopatinf` is true then no other scenario with the same ancestor is run. Note that since the order in which the different scenarios is run is not deterministic, this might introduce nondeterminism even if the sampling is deterministic.
"""
function iteration!(g::AbstractSDDPGraph, Ktot::Int, num_stages, verbose, pathsampler; ztol=1e-6, stopatinf=false, mergepaths=true, forwardcuts=false, backwardcuts=true)

pathsd, mastersol, stats = forwardpass!(g, Ktot, num_stages, verbose, pathsampler; ztol=ztol, stopatinf=stopatinf, mergepaths=mergepaths, forwardcuts=forwardcuts)

if backwardcuts
backwardpass!(g, num_stages, pathsd, pathsampler, stats; ztol=ztol)
end

mastersol, stats
end

Expand All @@ -145,7 +157,7 @@ function SDDP(g::AbstractSDDPGraph, num_stages; K::Int=25, stopcrit::AbstractSto
stats.niterations = 1

while (mastersol === nothing || mastersol.status != :Infeasible) && !stop(stopcrit, stats, totalstats)
itertime = @_time mastersol, stats = iteration(g, K, num_stages, verbose, pathsampler, ztol=1e-6, stopatinf=stopatinf, mergepaths=mergepaths, forwardcuts=forwardcuts, backwardcuts=backwardcuts)
itertime = @_time mastersol, stats = iteration!(g, K, num_stages, verbose, pathsampler, ztol=1e-6, stopatinf=stopatinf, mergepaths=mergepaths, forwardcuts=forwardcuts, backwardcuts=backwardcuts)
stats.time = itertime

totalstats += stats
Expand Down