Skip to content

Commit

Permalink
stability
Browse files Browse the repository at this point in the history
  • Loading branch information
annamariadziubyna committed Feb 25, 2024
1 parent 0ae6297 commit de8e175
Show file tree
Hide file tree
Showing 2 changed files with 9,444 additions and 0 deletions.
160 changes: 160 additions & 0 deletions benchmarks/stability.jl
Original file line number Diff line number Diff line change
@@ -0,0 +1,160 @@
# using MPI
using LinearAlgebra
using MKL
using SpinGlassEngine
using SpinGlassNetworks
using SpinGlassTensors
#using SpinGlassExhaustive
using Logging
using CSV
using DataFrames
using Memoization
using JSON3

#=
function brute_force_gpu(ig::IsingGraph; num_states::Int)
brute_force(ig, :GPU, num_states=num_states)
end
=#

# MPI.Init()
size = 1
rank = 0
# size = MPI.Comm_size(MPI.COMM_WORLD)
# rank = MPI.Comm_rank(MPI.COMM_WORLD)


M, N, T = 7,7,3
INSTANCE_DIR = "$(@__DIR__)/../test/instances/pegasus_random/P8/RCO/SpinGlass/single"
OUTPUT_DIR = "$(@__DIR__)/results/pegasus_random/P8/RCO/droplets/final_bench_Z_EngGauges"

if !Base.Filesystem.isdir(OUTPUT_DIR)
Base.Filesystem.mkpath(OUTPUT_DIR)
end

BETAS = collect(0.1:0.1:1.0) #[0.5,]
LAYOUT = (EnergyGauges,)
TRANSFORM = [all_lattice_transformations[4],]

GAUGE = NoUpdate
STRATEGY = Zipper
SPARSITY = Sparse
graduate_truncation = :graduate_truncate

INDβ = [3,]
MAX_STATES = [1024,]
BOND_DIM = [8,]

MAX_SWEEPS = [0,]
VAR_TOL = 1E-16
TOL_SVD = 1E-16
ITERS_SVD = 2
ITERS_VAR = 1
DTEMP_MULT = 2
METHOD = :psvd_sparse
I = [1,]
eng = [40, ]
hamming_dist = 74

disable_logging(LogLevel(1))
# BLAS.set_num_threads(1)

function pegasus_sim(inst, trans, β, Layout, bd, ms, eng, hamming_dist, mstates)
δp = 0.0

cl_h = clustered_hamiltonian(
ising_graph(INSTANCE_DIR * "/" * inst),
spectrum=full_spectrum,
cluster_assignment_rule=pegasus_lattice((M, N, T))
)

params = MpsParameters(bd, VAR_TOL, ms, TOL_SVD, ITERS_SVD, ITERS_VAR, DTEMP_MULT, METHOD)
search_params = SearchParameters(mstates, δp)

net = PEPSNetwork{SquareCrossDoubleNode{Layout}, SPARSITY}(M, N, cl_h, trans)
ctr = MpsContractor{STRATEGY, GAUGE}(net, [β/6, β/3, β/2, β], graduate_truncation, params)
sol1, schmidts = low_energy_spectrum(ctr, search_params, merge_branches(ctr, :nofit, SingleLayerDroplets(eng, hamming_dist, :hamming)))

sol2 = unpack_droplets(sol1, β)
ig_states = decode_clustered_hamiltonian_state.(Ref(cl_h), sol2.states)

ldrop = length(sol2.states)
cRAM = round(Base.summarysize(Memoization.caches) * 1E-9; sigdigits=2)
clear_memoize_cache()
sol1, ctr, cRAM, schmidts, ldrop, sol2, ig_states
end

function run_bench(inst::String, β::Real, t, l, bd, ms, eng, hamming_dist, mstates, i)
hash_name = hash(string(inst, β, t, l, bd, ms, eng, hamming_dist, mstates, i))
out_path = string(OUTPUT_DIR, "/", hash_name, ".json")

if isfile(out_path)
println("Skipping for , $t, $l, $bd, $eng, $hamming_dist, $ms, $mstates.")
else
data = try
tic_toc = @elapsed sol, ctr, cRAM, schmidts, ldrop, droplets, ig_states = pegasus_sim(inst, t, β, l, bd, ms, eng, hamming_dist, mstates)

data = DataFrame(
:instance => inst,
=> β,
:Layout => l,
:transform => t,
:energy => sol.energies[begin],
:probabilities => sol.probabilities,
:discarded_probability => sol.largest_discarded_probability,
:statistic => minimum(values(ctr.statistics)),
:max_states => mstates,
:bond_dim => bd,
:max_sweeps => ms,
:eng => eng,
:hamming_dist => hamming_dist,
:drop_eng => [droplets.energies],
:drop_states => [droplets.states],
:ig_states => [ig_states],
:drop_prob => [droplets.probabilities],
:drop_degeneracy => [droplets.degeneracy],
:drop_ldp => [droplets.largest_discarded_probability],
:drop_number => ldrop,
:iters_svd => ITERS_SVD,
:iters_var => ITERS_VAR,
:dtemp_mult => DTEMP_MULT,
:var_tol => VAR_TOL,
:time => tic_toc,
:cRAM => cRAM,
:schmidts => schmidts
)
catch err
data = DataFrame(
:instance => inst,
=> β,
:Layout => l,
:transform => t,
:max_states => mstates,
:hamming_dist => hamming_dist,
:iters_svd => ITERS_SVD,
:iters_var => ITERS_VAR,
:dtemp_mult => DTEMP_MULT,
:bond_dim => bd,
:max_sweeps => ms,
:var_tol => VAR_TOL,
:error => err
)
end

json_data = JSON3.write(data)
open(out_path, "w") do io
print(io, json_data)
end

end
end

all_params = collect(
Iterators.product(
readdir(INSTANCE_DIR, join=false), BETAS, TRANSFORM, LAYOUT, BOND_DIM, MAX_SWEEPS, eng, hamming_dist, MAX_STATES, I)
)

for i (1+rank):size:length(all_params)
run_bench(all_params[i]...)
GC.gc()
end
Loading

0 comments on commit de8e175

Please sign in to comment.