-
Notifications
You must be signed in to change notification settings - Fork 3
/
bar.py
135 lines (100 loc) · 3.65 KB
/
bar.py
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
from vmad import Builder, autooperator
from vmad.lib import fastpm, mpi, linalg
from vmad.contrib import cosmo4d
from abopt.abopt2 import TrustRegionCG
from abopt.abopt2 import LBFGS
import numpy
from nbodykit.cosmology import Planck15, LinearPower
pm = cosmo4d.ParticleMesh([8, 8, 8], BoxSize=16.)
def print(*args, **kwargs):
comm = pm.comm
from builtins import print
if comm.rank == 0:
print(*args, **kwargs)
def pprint(*args, **kwargs):
comm = pm.comm
from pprint import pprint
if comm.rank == 0:
pprint(*args, **kwargs)
Pss = LinearPower(Planck15, 0)
Pnn = lambda k: 1.0
ForwardModelHyperParameters = dict(
q = pm.generate_uniform_particle_grid(),
stages=[0.5, 1.0],
cosmology=Planck15,
pm=pm)
ForwardOperator = cosmo4d.FastPMOperator.bind(**ForwardModelHyperParameters)
def monitor(state):
#problem.save('/tmp/bar-%04d' % state['nit'], state)
z = state.g
print(abs(z.c2r().r2c() - z)[...].max() / z.cnorm())
print(state)
sim_t = cosmo4d.SynthData.create(ForwardOperator, 333, Pss, Pnn)
for i in range(3):
sim_b = cosmo4d.SynthData.create(ForwardOperator, 333, Pss, Pnn)
pprint(sim_t.attrs)
sim_t.save('/tmp/bar-truth')
sim_t = cosmo4d.SynthData.load('/tmp/bar-truth', pm.comm)
trcg = TrustRegionCG(
maxradius = 100000,
minradius = 1e-2,
initradus = 1,
atol = 0.1,
cg_rtol = 0.1,
cg_maxiter= 10,
)
trcg.cg_monitor = print
class MAPInversion(cosmo4d.MAPInversion):
def problem_factory(self, S, N, d, smoothing):
""" Create a problem object for a given set of args.
Parameters
----------
S : ComplexField
prior power
N : RealField
noise variance
d : RealField
data
smoothing : float
subsampling fraction (length scale of a gaussian smoothing)
"""
pm = self.ForwardOperator.hyperargs['pm']
problem = cosmo4d.ChiSquareProblem(pm.comm,
self.ForwardOperator,
[
cosmo4d.PriorOperator.bind(invS=S ** -1),
cosmo4d.SmoothedNLResidualOperator.bind(d=d, invN=N ** -1, scale=smoothing),
]
)
return problem
def schedule(self, S, N):
self.schedule_problem('S', S)
self.schedule_problem('N', N)
self.schedule_optimizer('maxiter', [1, 2, 3, 4])
self.schedule_problem('smoothing', [2, 1, 0, 0])
mapinv = MAPInversion(trcg, ForwardOperator)
# checking the problem
problem = mapinv.problem_factory(S=sim_t.S, N=sim_t.N, d=sim_t.d, smoothing=0)
print('objective(truth) =', problem.f(sim_t.s), 'expecting', pm.Nmesh.prod() * len(problem.residuals))
print('objective(0) =', problem.f(sim_t.s * 0.001))
mapinv.schedule(S=sim_t.S, N=sim_t.N)
shat_t = mapinv.apply(sim_t.d, epochs=[0, 1, 2, 3],
monitor_epoch=print,
monitor_progress=monitor)
sims = [None] * 8
shats = [None] * 8
for i in range(8):
sims[i] = cosmo4d.SynthData.create(ForwardOperator, i, Pss, Pnn)
shats[i] = mapinv.apply(sims[i].d, epochs=[0, 1, 2, 3],
monitor_epoch=print,
monitor_progress=monitor)
#print('gradient = ', problem.g(t * 0.001))
#print('hessian vector product = ', problem.hessian_vector_product(x, x))
#print('hessian vector product = ', problem.hessian_vector_product(x, x).shape)
lbfgs = LBFGS(maxiter=3)
problem.set_preconditioner('complex')
s1 = sim_t.s * 0.001
#state = lbfgs.minimize(problem, t* 0.001, monitor=monitor)
#s1 = state['x']
#problem.set_preconditioner('real')
state = trcg.minimize(problem, s1, monitor=monitor)