-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathextra_model_workflow.jl
1272 lines (1144 loc) · 43.1 KB
/
extra_model_workflow.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
##########################################################################
# Benchmark for models that are not of the cnn type so they
# can not be compared to INS directly.
##########################################################################
#! format: off
if false #src
include("src/Benchmark.jl") #src
end #src
@info "Script started"
@info VERSION
using Pkg
@info Pkg.status()
# Color palette for consistent theme throughout paper
palette = (; color = ["#3366cc", "#cc0000", "#669900", "#ff9900"])
########################################################################## #src
# Read the configuration file
using IncompressibleNavierStokes
using NeuralClosure
using CoupledNODE
using Lux
using LuxCUDA
NS = Base.get_extension(CoupledNODE, :NavierStokes)
function load_config()
if haskey(ENV, "CONF_FILE")
@info "Reading configuration file from ENV:" ENV["CONF_FILE"]
return NS.read_config(ENV["CONF_FILE"])
elseif length(ARGS) > 0
@info "Reading configuration file from ARGS:" ARGS[1]
return NS.read_config(ARGS[1])
else
@info "Reading default configuration file: configs/conf_3.yaml"
return NS.read_config("configs/conf_3.yaml")
end
end
conf = load_config()
########################################################################## #src
# Choose where to put output
basedir = haskey(ENV, "DEEPDIP") ? ENV["DEEPDIP"] : @__DIR__
outdir = joinpath(basedir, "output", "kolmogorov")
closure_name = conf["closure"]["name"]
outdir_model = joinpath(outdir, closure_name)
plotdir = joinpath(outdir, "plots", closure_name)
logdir = joinpath(outdir, "logs", closure_name)
ispath(outdir) || mkpath(outdir)
ispath(outdir_model) || mkpath(outdir_model)
ispath(plotdir) || mkpath(plotdir)
ispath(logdir) || mkpath(logdir)
# Turn off plots for array jobs.
# If all the workers do this at the same time, one might
# error when saving the file at the same time
doplot() = true
########################################################################## #src
# ## Configure logger ``
using Benchmark
using Dates
# Write output to file, as the default SLURM file is not updated often enough
isslurm = haskey(ENV, "SLURM_JOB_ID")
if isslurm
jobid = parse(Int, ENV["SLURM_JOB_ID"])
taskid = parse(Int, ENV["SLURM_ARRAY_TASK_ID"])
numtasks = parse(Int, ENV["SLURM_ARRAY_TASK_COUNT"])
logfile = "job=$(jobid)_task=$(taskid)_$(Dates.now()).out"
else
taskid = 1
numtasks = 1
logfile = "log_$(Dates.now()).out"
end
logfile = joinpath(logdir, logfile)
setsnelliuslogger(logfile)
@info "# A-posteriori analysis: Forced turbulence (2D)"
# ## Load packages
@info "Loading packages"
if "CoupledNODE" in keys(Pkg.installed())
@info "CoupledNODE already installed"
else
Pkg.add(PackageSpec(rev = "main", url = "https://github.com/DEEPDIP-project/CoupledNODE.jl.git"))
end
using Accessors
using Adapt
using CairoMakie
using CoupledNODE: loss_priori_lux, create_loss_post_lux
using CUDA
using DifferentialEquations
using IncompressibleNavierStokes.RKMethods
using JLD2
using LaTeXStrings
using LinearAlgebra
using Lux
using LuxCUDA
using NNlib
using Optimisers
using ParameterSchedulers
using Random
using SparseArrays
# ## Random number seeds
#
# Use a new RNG with deterministic seed for each code "section"
# so that e.g. training batch selection does not depend on whether we
# generated fresh filtered DNS data or loaded existing one (the
# generation of which would change the state of a global RNG).
#
# Note: Using `rng = Random.default_rng()` twice seems to point to the
# same RNG, and mutating one also mutates the other.
# `rng = Xoshiro()` creates an independent copy each time.
#
# We define all the seeds here.
seeds = NS.load_seeds(conf)
########################################################################## #src
# ## Hardware selection
# Precision
T = eval(Meta.parse(conf["T"]))
# Device
if CUDA.functional()
## For running on a CUDA compatible GPU
@info "Running on CUDA"
cuda_active = true
backend = CUDABackend()
CUDA.allowscalar(false)
device = x -> adapt(CuArray, x)
@info "CUDA device" CUDA.device()
clean() = (GC.gc(); CUDA.reclaim())
else
## For running on CPU.
## Consider reducing the sizes of DNS, LES, and CNN layers if
## you want to test run on a laptop.
@warn "Running on CPU"
cuda_active = false
backend = CPU()
device = identity
clean() = nothing
end
conf["params"]["backend"] = deepcopy(backend)
@info backend
########################################################################## #src
# ## Data generation
#
# Create filtered DNS data for training, validation, and testing.
# Parameters
params = NS.load_params(conf)
# DNS seeds
ntrajectory = conf["ntrajectory"]
dns_seeds = splitseed(seeds.dns, ntrajectory)
dns_seeds_train = dns_seeds[1:ntrajectory-2]
dns_seeds_valid = dns_seeds[ntrajectory-1:ntrajectory-1]
dns_seeds_test = dns_seeds[ntrajectory:ntrajectory]
# Create data
docreatedata = conf["docreatedata"]
for i = 1:ntrajectory
if i%numtasks == taskid - 1
docreatedata && createdata(; params, seed = dns_seeds[i], outdir, backend)
end
end
@info "Data generated"
# Computational time
docomp = conf["docomp"]
docomp && let
comptime, datasize = 0.0, 0.0
for seed in dns_seeds
comptime += load(
getdatafile(outdir, params.nles[1], params.filters[1], seed),
"comptime",
)
end
for seed in dns_seeds, nles in params.nles, Φ in params.filters
data = namedtupleload(getdatafile(outdir, nles, Φ, seed))
datasize += Base.summarysize(data)
end
@info "Data" comptime
@info "Data" comptime / 60 datasize * 1e-9
clean()
end
# LES setups
setups = map(nles -> getsetup(; params, nles), params.nles);
########################################################################## #src
# ## CNN closure model
# All training sessions will start from the same θ₀
# for a fair comparison.
using Lux:relu
# Load the modules for AttentionCNN
using AttentionLayer
using CoupledNODE:Base, AttentionCNN
ACNN = Base.get_extension(CoupledNODE, :AttentionCNN)
# Load the modules for CNO
using ConvolutionalNeuralOperators
using CoupledNODE:Base, CNO
CNO = Base.get_extension(CoupledNODE, :CNO)
closure, θ_start, st = NS.load_model(conf)
@info "Initialized CNN with $(length(θ_start)) parameters"
# Give the CNN a test run
# Note: Data and parameters are stored on the CPU, and
# must be moved to the GPU before use (with `device`)
let
@info "CNN warm up run"
using NeuralClosure.Zygote
u = randn(T, 32+2, 32+2, 2, 10) |> device
θ = θ_start |> device
closure(u, θ, st)
gradient(θ -> sum(closure(u, θ, st)[1]), θ)
clean()
end
########################################################################## #src
# ## Training
# ### A-priori training
#
# Train one set of CNN parameters for each of the filter types and grid sizes.
# Use the same batch selection random seed for each training setup.
# Save parameters to disk after each run.
# Plot training progress (for a validation data batch).
# Train
@info "A priori training"
for i = 1:ntrajectory
if i%numtasks == taskid -1
let
dotrain = conf["priori"]["dotrain"]
nepoch = conf["priori"]["nepoch"]
dotrain && trainprior(;
params,
priorseed = seeds.prior,
dns_seeds_train,
dns_seeds_valid,
taskid = i,
outdir,
plotdir,
closure,
closure_name,
θ_start,
st,
opt = eval(Meta.parse(conf["priori"]["opt"])),
batchsize = conf["priori"]["batchsize"],
do_plot = conf["priori"]["do_plot"],
plot_train = conf["priori"]["plot_train"],
nepoch,
)
end
end
end
# Load learned parameters and training times
priortraining = loadprior(outdir, closure_name, params.nles, params.filters)
θ_cnn_prior = map(p -> copyto!(copy(θ_start), p.θ), priortraining)
@info "" θ_cnn_prior .|> extrema # Check that parameters are within reasonable bounds
# Training times
map(p -> p.comptime, priortraining)
map(p -> p.comptime, priortraining) |> vec .|> x -> round(x; digits = 1)
map(p -> p.comptime, priortraining) |> sum |> x -> x / 60 # Minutes
# ## Plot training history
with_theme(; palette) do
doplot() || return
fig = Figure(; size = (950, 250))
for (ig, nles) in enumerate(params.nles)
ax = Axis(
fig[1, ig];
title = "n = $(nles)",
xlabel = "Iteration",
ylabel = "A-priori error",
ylabelvisible = ig == 1,
yticksvisible = ig == 1,
yticklabelsvisible = ig == 1,
)
ylims!(-0.05, 1.05)
lines!(
ax,
# [Point2f(0, 1), Point2f(priortraining[ig, 1].lhist_nomodel[end][1], 1)];
priortraining[ig, 1].lhist_nomodel,
label = "No closure",
linestyle = :dash,
)
for (ifil, Φ) in enumerate(params.filters)
label = Φ isa FaceAverage ? "FA" : "VA"
lines!(ax, priortraining[ig, ifil].lhist_val; label)
end
end
axes = filter(x -> x isa Axis, fig.content)
linkaxes!(axes...)
Legend(fig[1, end+1], axes[1])
figdir = joinpath(plotdir, "priortraining")
ispath(figdir) || mkpath(figdir)
save("$figdir/validationerror.pdf", fig)
display(fig)
end
########################################################################## #src
# ### A-posteriori training
#
# Train one set of CNN parameters for each
# projection order, filter type and grid size.
# Use the same batch selection random seed for each training setup.
# Save parameters to disk after each combination.
# Plot training progress (for a validation data batch).
#
# [INS] The time stepper `RKProject` allows for choosing when to project.
# [CNODE] Only DCF (last) is supported since it appears to be the best one.
projectorders = eval(Meta.parse(conf["posteriori"]["projectorders"]))
nprojectorders = length(projectorders)
@assert nprojectorders == 1 "Only DCF should be done"
# Train
for i = 1:ntrajectory
if i%numtasks == taskid -1
let
dotrain = conf["posteriori"]["dotrain"]
nepoch = conf["posteriori"]["nepoch"]
dotrain && trainpost(;
params,
projectorders,
outdir,
plotdir,
taskid = i,
postseed = seeds.post,
dns_seeds_train,
dns_seeds_valid,
nunroll = conf["posteriori"]["nunroll"],
closure,
closure_name,
θ_start = θ_cnn_prior,
st,
opt = eval(Meta.parse(conf["posteriori"]["opt"])),
nunroll_valid = conf["posteriori"]["nunroll_valid"],
nepoch,
dt = eval(Meta.parse(conf["posteriori"]["dt"])),
do_plot = conf["posteriori"]["do_plot"],
plot_train = conf["posteriori"]["plot_train"],
)
end
end
end
# Load learned parameters and training times
posttraining = loadpost(outdir, closure_name, params.nles, params.filters, projectorders)
θ_cnn_post = map(p -> copyto!(copy(θ_start), p.θ), posttraining)
@info "" θ_cnn_post .|> extrema # Check that parameters are within reasonable bounds
# Training times
map(p -> p.comptime, posttraining) ./ 60
map(p -> p.comptime, posttraining) |> sum |> x -> x / 60
map(p -> p.comptime, posttraining) |> x -> reshape(x, :, nprojectorders) .|> x -> round(x; digits = 1)
# ## Plot a-posteriori training history
with_theme(; palette) do
doplot() || return
fig = Figure(; size = (950, 400))
for (iorder, projectorder) in enumerate(projectorders)
axes = []
for (ig, nles) in enumerate(params.nles)
ax = Axis(
fig[iorder, ig];
title = "n = $(nles)",
xlabel = "Iteration",
ylabel = projectorder == ProjectOrder.First ? "DIF" : "DCF",
ylabelvisible = ig == 1,
ylabelfont = :bold,
titlevisible = iorder == 1,
xlabelvisible = iorder == 2,
xticksvisible = iorder == 2,
xticklabelsvisible = iorder == 2,
)
for (ifil, Φ) in enumerate(params.filters)
postfile = Benchmark.getpostfile(outdir, closure_name, nles, Φ, projectorder)
check = namedtupleload(postfile)
# print the keys of the checkpoint
(; lhist_val) = check[1]
label = Φ isa FaceAverage ? "FA" : "VA"
lines!(ax, lhist_val; color = Cycled(ifil + 1), label)
end
ig == 4 && iorder == 1 && axislegend(ax)
push!(axes, ax)
end
# linkaxes!(axes...)
end
# axes = filter(x -> x isa Axis, fig.content)
# linkaxes!(axes...)
# Legend(fig[:, end+1], filter(x -> x isa Axis, fig.content)[1])
Label(fig[0, :], "A-posteriori error"; valign = :bottom, font = :bold)
rowgap!(fig.layout, 10)
figdir = joinpath(plotdir, "posttraining")
ispath(figdir) || mkpath(figdir)
save("$figdir/validationerror.pdf", fig)
display(fig)
end
########################################################################## #src
# ## Prediction errors
# ### Compute a-priori errors
#
# Note that it is still interesting to compute the a-priori errors for the
# a-posteriori trained CNN.
let
eprior = (;
nomodel = ones(T, length(params.nles)),
prior = zeros(T, size(θ_cnn_prior)),
post = zeros(T, size(θ_cnn_post)),
)
for (ifil, Φ) in enumerate(params.filters), (ig, nles) in enumerate(params.nles)
@info "Computing a-priori errors" Φ nles
setup = getsetup(; params, nles)
data = map(s -> namedtupleload(getdatafile(outdir, nles, Φ, s)), dns_seeds_test)
testset = create_io_arrays(data, setup)
i = 1:min(1000, size(testset.u, 4))
u, c = testset.u[:, :, :, i], testset.c[:, :, :, i]
testset = (u, c) |> device
priori_err(θ) = loss_priori_lux(closure, θ, st, testset)
eprior.prior[ig, ifil] = priori_err(device(θ_cnn_prior[ig, ifil]))[1]
for iorder in eachindex(projectorders)
eprior.post[ig, ifil, iorder] = priori_err(device(θ_cnn_post[ig, ifil, iorder]))[1]
end
end
jldsave(joinpath(outdir_model, "eprior.jld2"); eprior...)
end
clean()
eprior = namedtupleload(joinpath(outdir_model, "eprior.jld2"))
########################################################################## #src
# ### Compute a-posteriori errors
let
sample = namedtupleload(
getdatafile(outdir, params.nles[1], params.filters[1], dns_seeds_test[1]),
)
sample.t[100]
end
let
s = (length(params.nles), length(params.filters), length(projectorders))
epost = (;
nomodel = zeros(T, s),
cnn_prior = zeros(T, s),
cnn_post = zeros(T, s),
)
for (iorder, projectorder) in enumerate(projectorders),
(ifil, Φ) in enumerate(params.filters),
(ig, nles) in enumerate(params.nles)
@info "Computing a-posteriori errors" projectorder Φ nles
I = CartesianIndex(ig, ifil, iorder)
setup = getsetup(; params, nles)
psolver = psolver_spectral(setup)
sample = namedtupleload(getdatafile(outdir, nles, Φ, dns_seeds_test[1]))
it = 1:100
data = (;
u = selectdim(sample.u, ndims(sample.u), it) |> collect |> device,
t = sample.t[it],
)
dt = T(1e-3)
## No model
dudt_nomod = NS.create_right_hand_side(
setup, psolver)
err_post = create_loss_post_lux(dudt_nomod; sciml_solver = Tsit5(), dt = dt, use_cuda = CUDA.functional())
epost.nomodel[I] = err_post(closure, θ_cnn_post[I].*0 , st, data)[1]
# with closure
dudt = NS.create_right_hand_side_with_closure(
setup, psolver, closure, st)
err_post = create_loss_post_lux(dudt; sciml_solver = Tsit5(), dt = dt, use_cuda = CUDA.functional())
epost.cnn_prior[I] = err_post(closure, device(θ_cnn_prior[ig, ifil]), st, data)[1]
epost.cnn_post[I] = err_post(closure, device(θ_cnn_post[I]), st, data)[1]
clean()
end
jldsave(joinpath(outdir_model, "epost.jld2"); epost...)
end
epost = namedtupleload(joinpath(outdir_model, "epost.jld2"))
########################################################################## #src
# ### Plot a-priori errors
# Better for PDF export
CairoMakie.activate!()
with_theme(; palette) do
fig = Figure(; size = (800, 300))
axes = []
for (ifil, Φ) in enumerate(params.filters)
ax = Axis(
fig[1, ifil];
xscale = log10,
xticks = params.nles,
xlabel = "Resolution",
# title = "Relative a-priori error $(ifil == 1 ? " (FA)" : " (VA)")",
# title = "$(Φ isa FaceAverage ? "FA" : "VA")",
title = "$(Φ isa FaceAverage ? "Face-averaging" : "Volume-averaging")",
ylabel = "A-priori error",
ylabelvisible = ifil == 1,
yticksvisible = ifil == 1,
yticklabelsvisible = ifil == 1,
)
for (e, marker, label, color) in [
(eprior.nomodel, :circle, "No closure", Cycled(1)),
(eprior.prior[:, ifil], :utriangle, "CNN (prior)", Cycled(2)),
(eprior.post[:, ifil, 1], :diamond, "CNN (post, DCF)", Cycled(3)),
]
scatterlines!(params.nles, e; marker, color, label)
end
# axislegend(; position = :lb)
ylims!(ax, (T(-0.05), T(1.05)))
push!(axes, ax)
end
Legend(fig[1, end+1], axes[1])
save("$plotdir/eprior.pdf", fig)
display(fig)
end
########################################################################## #src
# ### Plot a-posteriori errors
# Better for PDF export
CairoMakie.activate!()
with_theme(; palette) do
doplot() || return
fig = Figure(; size = (800, 300))
linestyles = [:solid, :dash]
linestyles = [:solid]
for (iorder, projectorder) in enumerate(projectorders)
lesmodel = iorder == 1 ? "DCF" : "DCF"
(; nles) = params
ax = Axis(
fig[1, iorder];
xscale = log10,
yscale = log10,
xticks = nles,
xlabel = "Resolution",
title = "$lesmodel",
ylabel = "A-posteriori error",
ylabelvisible = iorder == 1,
yticksvisible = iorder == 1,
yticklabelsvisible = iorder == 1,
)
for (e, marker, label, color) in [
(epost.nomodel, :circle, "No closure", Cycled(1)),
(epost.cnn_prior, :rect, "CNN (Lprior)", Cycled(3)),
(epost.cnn_post, :diamond, "CNN (Lpost)", Cycled(4)),
]
for (ifil, linestyle) in enumerate(linestyles)
ifil == 2 && (label = nothing)
scatterlines!(nles, e[:, ifil, iorder]; color, linestyle, marker, label)
end
end
# ylims!(ax, (T(0.025), T(1.00)))
end
linkaxes!(filter(x -> x isa Axis, fig.content)...)
g = GridLayout(fig[1, end+1])
Legend(g[1, 1], filter(x -> x isa Axis, fig.content)[1]; valign = :bottom)
Legend(
g[2, 1],
map(s -> LineElement(; color = :black, linestyle = s), linestyles),
["FA"];
orientation = :horizontal,
valign = :top,
)
rowsize!(g, 1, Relative(1 / 2))
save("$plotdir/epost.pdf", fig)
display(fig)
end
########################################################################## #src
# ## Energy evolution
# ### Compute total kinetic energy as a function of time
CUDA.allowscalar() do
let
s = length(params.nles), length(params.filters), length(projectorders)
keys = [:ref, :nomodel, :cnn_prior, :cnn_post]
divergencehistory = (; map(k -> k => fill(Point2f[], s), keys)...)
energyhistory = (; map(k -> k => fill(Point2f[], s), keys)...)
for (iorder, projectorder) in enumerate(projectorders),
(ifil, Φ) in enumerate(params.filters),
(ig, nles) in enumerate(params.nles)
I = CartesianIndex(ig, ifil, iorder)
@info "Computing divergence and kinetic energy" projectorder Φ nles
setup = getsetup(; params, nles)
psolver = default_psolver(setup)
sample = namedtupleload(getdatafile(outdir, nles, Φ, dns_seeds_test[1]))
ustart = selectdim(sample.u, ndims(sample.u), 1) |> collect |> device
T = eltype(ustart)
# Shorter time for DIF
t_DIF = T(1)
# Reference trajectories
divergencehistory.ref[I] = let
div = scalarfield(setup)
udev = vectorfield(setup)
it = 1:5:length(sample.t)
map(it) do it
t = sample.t[it]
u = selectdim(sample.u, ndims(sample.u), it)
copyto!(udev, u)
IncompressibleNavierStokes.divergence!(div, udev, setup)
d = view(div, setup.grid.Ip)
d = sum(abs2, d) / length(d)
d = sqrt(d)
Point2f(t, d)
end
end
energyhistory.ref[I] = let
it = 1:5:length(sample.t)
udev = vectorfield(setup)
map(it) do it
t = sample.t[it]
u = selectdim(sample.u, ndims(sample.u), it)
copyto!(udev, u)
Point2f(t, total_kinetic_energy(udev, setup))
end
end
nupdate = 5
writer = processor() do state
div = scalarfield(setup)
dhist = Point2f[]
ehist = zeros(Point2f, 0)
on(state) do (; u, t, n)
if n % nupdate == 0
IncompressibleNavierStokes.divergence!(div, u, setup)
d = view(div, setup.grid.Ip)
d = sum(abs2, d) / length(d)
d = sqrt(d)
push!(dhist, Point2f(t, d))
e = total_kinetic_energy(u, setup)
push!(ehist, Point2f(t, e))
end
end
state[] = state[] # Compute initial divergence
(; dhist, ehist)
end
for (sym, closure_model, θ) in [
(:nomodel, nothing, nothing),
(:cnn_prior, wrappedclosure(closure_INS, setup), device(θ_cnn_prior[ig, ifil])),
(:cnn_post, wrappedclosure(closure_INS, setup), device(θ_cnn_post[I])),
]
_, results = solve_unsteady(;
setup = (; setup..., closure_model),
ustart,
tlims = (
sample.t[1],
projectorder == ProjectOrder.First ? t_DIF : sample.t[end],
),
Δt_min = T(1e-5),
method = RKProject(params.method, projectorder),
processors = (; writer, logger = timelogger(; nupdate = 500)),
psolver,
θ,
)
divergencehistory[sym][I] = results.writer.dhist
energyhistory[sym][I] = results.writer.ehist
end
end
jldsave(joinpath(outdir_model, "history.jld2"); energyhistory, divergencehistory)
clean()
end
end
(; divergencehistory, energyhistory) = namedtupleload(joinpath(outdir, "history.jld2"));
########################################################################## #src
# Check that energy is within reasonable bounds
energyhistory.ref .|> extrema
energyhistory.nomodel .|> extrema
energyhistory.cnn_prior .|> extrema
energyhistory.cnn_post .|> extrema
# Check that divergence is within reasonable bounds
divergencehistory.ref .|> extrema
divergencehistory.nomodel .|> extrema
divergencehistory.cnn_prior .|> extrema
divergencehistory.cnn_post .|> extrema
########################################################################## #src
# ### Plot energy evolution
# Better for PDF export
CairoMakie.activate!()
with_theme(; palette) do
doplot() || return
for (igrid, nles) in enumerate(params.nles)
@info "Plotting energy evolution" nles
fig = Figure(; size = (800, 450))
g = GridLayout(fig[1, 1])
for (iorder, projectorder) in enumerate(projectorders),
(ifil, Φ) in enumerate(params.filters)
I = CartesianIndex(igrid, ifil, iorder)
subfig = g[ifil, iorder]
ax = Axis(
subfig;
# xscale = log10,
# yscale = log10,
xlabel = "t",
# ylabel = Φ isa FaceAverage ? "Face-average" : "Volume-average",
ylabel = "E(t)",
# ylabelfont = :bold,
title = projectorder == ProjectOrder.First ? "DIF" : "DCF",
titlevisible = ifil == 1,
xlabelvisible = ifil == 2,
xticksvisible = ifil == 2,
xticklabelsvisible = ifil == 2,
ylabelvisible = iorder == 1,
yticksvisible = iorder == 1,
yticklabelsvisible = iorder == 1,
aspect = DataAspect(),
)
# xlims!(ax, (1e-2, 5.0))
# xlims!(ax, (0.0, 1.0))
# ylims!(ax, (1.3, 2.3))
plots = [
(energyhistory.nomodel, :solid, 1, "No closure"),
(energyhistory.cnn_prior, :solid, 3, "CNN (prior)"),
(energyhistory.cnn_post, :solid, 4, "CNN (post)"),
(energyhistory.ref, :dash, 1, "Reference"),
]
for (p, linestyle, i, label) in plots
lines!(ax, p[I]; color = Cycled(i), linestyle, label)
iorder == 1 && xlims!(-0.05, 1.05)
# iorder == 1 && ylims!(1.1, 3.1)
ylims!(1.3, 3.0)
end
# Plot zoom-in box
if iorder == 2
tlims = iorder == 1 ? (0.05, 0.2) : (0.8, 1.2)
i1 = findfirst(p -> p[1] > tlims[1], energyhistory.ref[I])
i2 = findfirst(p -> p[1] > tlims[2], energyhistory.ref[I])
tlims = energyhistory.ref[I][i1][1], energyhistory.ref[I][i2][1]
klims = energyhistory.ref[I][i1][2], energyhistory.ref[I][i2][2]
dk = klims[2] - klims[1]
# klims = klims[1] - 0.2 * dk, klims[2] + 0.2 * dk
w = iorder == 1 ? 0.2 : 0.1
klims = klims[1] + w * dk, klims[2] - w * dk
box = [
Point2f(tlims[1], klims[1]),
Point2f(tlims[2], klims[1]),
Point2f(tlims[2], klims[2]),
Point2f(tlims[1], klims[2]),
Point2f(tlims[1], klims[1]),
]
lines!(ax, box; color = :black)
ax2 = Axis(
subfig;
# bbox = BBox(0.8, 0.9, 0.2, 0.3),
width = Relative(0.35),
height = Relative(0.35),
halign = 0.05,
valign = 0.95,
limits = (tlims..., klims...),
xscale = log10,
yscale = log10,
xticksvisible = false,
xticklabelsvisible = false,
xgridvisible = false,
yticksvisible = false,
yticklabelsvisible = false,
ygridvisible = false,
backgroundcolor = :white,
)
# https://discourse.julialang.org/t/makie-inset-axes-and-their-drawing-order/60987/5
translate!(ax2.scene, 0, 0, 10)
translate!(ax2.elements[:background], 0, 0, 9)
for (p, linestyle, i, label) in plots
lines!(ax2, p[igrid, ifil, iorder]; color = Cycled(i), linestyle, label)
end
end
Label(
g[ifil, 0],
# Φ isa FaceAverage ? "Face-average" : "Volume-average";
Φ isa FaceAverage ? "FA" : "VA";
# halign = :right,
font = :bold,
# rotation = pi/2,
tellheight = false,
)
end
colgap!(g, 10)
rowgap!(g, 10)
# colsize!(g, 1, Relative(1 / 5))
Legend(fig[:, end+1], filter(x -> x isa Axis, fig.content)[1])
name = "$plotdir/energy_evolution/"
ispath(name) || mkpath(name)
save("$(name)/nles=$(nles).pdf", fig)
display(fig)
end
end
########################################################################## #src
# ### Plot Divergence
# Better for PDF export
CairoMakie.activate!()
with_theme(; palette) do
doplot() || return
islog = true
for (igrid, nles) in enumerate(params.nles)
@info "Plotting divergence" nles
fig = Figure(; size = (800, 450))
for (iorder, projectorder) in enumerate(projectorders),
(ifil, Φ) in enumerate(params.filters)
I = CartesianIndex(igrid, ifil, iorder)
subfig = fig[ifil, iorder]
ax = Axis(
subfig;
yscale = islog ? log10 : identity,
xlabel = "t",
ylabel = Φ isa FaceAverage ? "Face-average" : "Volume-average",
ylabelfont = :bold,
title = projectorder == ProjectOrder.First ? "DIF" : "DCF",
titlevisible = ifil == 1,
xlabelvisible = ifil == 2,
xticksvisible = ifil == 2,
xticklabelsvisible = ifil == 2,
ylabelvisible = iorder == 1,
yticksvisible = iorder == 1,
yticklabelsvisible = iorder == 1,
)
lines!(ax, divergencehistory.nomodel[I]; label = "No closure")
lines!(ax, divergencehistory.cnn_prior[I]; label = "CNN (prior)")
lines!(ax, divergencehistory.cnn_post[I]; label = "CNN (post)")
lines!(
ax,
divergencehistory.ref[I];
color = Cycled(1),
linestyle = :dash,
label = "Reference",
)
islog && ylims!(ax, (T(1e-6), T(1e3)))
iorder == 1 && xlims!(ax, (-0.05, 1.05))
end
rowgap!(fig.layout, 10)
colgap!(fig.layout, 10)
Legend(fig[:, end+1], filter(x -> x isa Axis, fig.content)[1])
name = "$plotdir/divergence/"
ispath(name) || mkpath(name)
save("$(name)/nles=$(nles).pdf", fig)
display(fig)
end
end
########################################################################## #src
# ## Solutions at different times
let
s = length(params.nles), length(params.filters), length(projectorders)
temp = zeros(T, ntuple(Returns(0), params.D + 1))
keys = [:ref, :nomodel, :cnn_prior, :cnn_post]
times = T[0.1, 0.5, 1.0, 5.0]
itime_max_DIF = 3
times_exact = copy(times)
utimes = map(t -> (; map(k -> k => fill(temp, s), keys)...), times)
for (iorder, projectorder) in enumerate(projectorders),
(ifil, Φ) in enumerate(params.filters),
(igrid, nles) in enumerate(params.nles)
@info "Computing test solutions" projectorder Φ nles
I = CartesianIndex(igrid, ifil, iorder)
setup = getsetup(; params, nles)
psolver = default_psolver(setup)
sample = namedtupleload(getdatafile(outdir, nles, Φ, dns_seeds_test[1]))
ustart = selectdim(sample.u, ndims(sample.u), 1) |> collect
t = sample.t
function solve(ustart, tlims, closure_model, θ)
result = solve_unsteady(;
setup = (; setup..., closure_model),
ustart = device(ustart),
tlims,
method = RKProject(params.method, projectorder),
psolver,
θ,
)[1].u |> Array
#@info result
#Array(result)
end
t1 = t[1]
for i in eachindex(times)
# Only first times for First
i > itime_max_DIF && projectorder == ProjectOrder.First && continue
# Time interval
t0 = t1
t1 = times[i]
# Adjust t1 to be exactly on a reference time
it = findfirst(>(t1), t)
if isnothing(it)
# Not found: Final time
it = length(t)
end
t1 = t[it]
tlims = (t0, t1)
times_exact[i] = t1
getprev(i, sym) = i == 1 ? ustart : utimes[i-1][sym][I]
# Compute fields
utimes[i].ref[I] = selectdim(sample.u, ndims(sample.u), it) |> collect
utimes[i].nomodel[I] = solve(getprev(i, :nomodel), tlims, nothing, nothing)
#utimes[i].nomodel[I,:] = solve(getprev(i, :nomodel), tlims, nothing, nothing)
utimes[i].cnn_prior[I] = solve(
getprev(i, :cnn_prior),
tlims,
wrappedclosure(closure_INS, setup),
device(θ_cnn_prior[igrid, ifil]),
)
utimes[i].cnn_post[I] = solve(
getprev(i, :cnn_post),
tlims,
wrappedclosure(closure_INS, setup),
device(θ_cnn_post[I]),
)
end
clean()
end
jldsave("$outdir_model/solutions.jld2"; u = utimes, t = times_exact, itime_max_DIF)
end;
# Load solution
solutions = namedtupleload("$outdir_model/solutions.jld2");
########################################################################## #src
# ### Plot spectra
#
# Plot kinetic energy spectra.
with_theme(; palette) do
doplot() || return
for (ifil, Φ) in enumerate(params.filters), (igrid, nles) in enumerate(params.nles)
@info "Plotting spectra" Φ nles
fig = Figure(; size = (800, 450))
fil = Φ isa FaceAverage ? "face-average" : "volume-average"
setup = getsetup(; params, nles)
(; Ip) = setup.grid
(; inds, κ, K) = IncompressibleNavierStokes.spectral_stuff(setup)