-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval_checkerboard_runs.py
204 lines (183 loc) · 9.82 KB
/
eval_checkerboard_runs.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
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
import os
from mpi4py import MPI
comm = MPI.COMM_WORLD
n_ranks = comm.Get_size()
rank = comm.Get_rank()
# supress warning about no cuda mpi version
# we don't need that because jax handles that, we only want to run copies of a process with some communication
os.environ["MPI4JAX_USE_CUDA_MPI"] = "0"
# set only one visible device
os.environ["CUDA_VISIBLE_DEVICES"] = f"{rank}"
# force to use gpu
os.environ["JAX_PLATFORM_NAME"] = "gpu"
# make netket compute the exact autocorrelation time
# os.environ["NETKET_EXPERIMENTAL_FFT_AUTOCORRELATION"] = "1"
# make netket use the split-Rhat diagnostic, not just the plain one
# os.environ["NETKET_USE_PLAIN_RHAT"] = "0"
import jax
from jax import numpy as jnp
import netket as nk
from netket.utils import HashableArray
import flax
import numpy as np
import pandas as pd
from tqdm import tqdm
import matplotlib
from matplotlib import pyplot as plt
import geneqs
from global_variables import RESULTS_PATH
cmap = matplotlib.colormaps["Set1"]
f_dict = {0: "x", 1: "y", 2: "z"}
L = 8
label = "hx_left_right_chaintransfer_nonoise"
save_dir = f"{RESULTS_PATH}/checkerboard/L={L}_final/L={L}_mc_crbm_{label}"
# %%
shape = jnp.array([L, L, L])
hilbert = nk.hilbert.Spin(s=1 / 2, N=jnp.prod(shape).item())
eval_model = "CheckerCRBM_2"
# get fields
direction_index = 0 # 0 for x, 1 for y, 2 for z;
obs = pd.read_csv(f"{save_dir}/L{shape}_{eval_model}_observables.txt", sep=" ", header=0)
field_strengths = obs.iloc[:, :3].values[8:]
print(field_strengths)
# hist_fields = np.array([[0.20, 0, 0],
# [0.40, 0, 0],
# [0.44, 0, 0],
# [0.45, 0, 0],
# [0.50, 0, 0],
# [0.70, 0, 0]])
n_chains = 256 * 4 # total number of MCMC chains, when runnning on GPU choose ~O(1000)
chunk_size = 1 * 1024
n_samples = 16 * 1024 * 24
n_discard_per_chain = 0
n_bins = 20
# %% define observables
if direction_index == 0:
abs_magnetization = geneqs.operators.observables.AbsXMagnetization(hilbert)
magnetization = 1 / hilbert.size * sum([nk.operator.spin.sigmax(hilbert, i) for i in range(hilbert.size)])
elif direction_index == 1:
abs_magnetization = geneqs.operators.observables.AbsYMagnetization(hilbert)
magnetization = 1 / hilbert.size * sum([nk.operator.spin.sigmay(hilbert, i) for i in range(hilbert.size)])
elif direction_index == 2:
abs_magnetization = geneqs.operators.observables.AbsZMagnetization(hilbert)
magnetization = 1 / hilbert.size * sum([nk.operator.spin.sigmaz(hilbert, i) for i in range(hilbert.size)])
positions = jnp.asarray([[x, y, z]
for x in range(shape[0])
for y in range(shape[1])
for z in range(shape[2])
if (x + y + z) % 2 == 0])
xcubes = nk.operator.LocalOperator(hilbert, dtype=complex)
zcubes = nk.operator.LocalOperator(hilbert, dtype=complex)
for pos in positions:
xcubes += geneqs.operators.checkerboard.get_netket_xcube(hilbert, pos, shape)
zcubes += geneqs.operators.checkerboard.get_netket_zcube(hilbert, pos, shape)
# %% create model and update rule
perms = geneqs.utils.indexing.get_translations_cubical3d(shape, shift=2)
perms = nk.utils.HashableArray(perms.astype(int))
# noinspection PyArgumentList
correlators = (HashableArray(geneqs.utils.indexing.get_cubes_cubical3d(shape, 2)),
HashableArray(geneqs.utils.indexing.get_bonds_cubical3d(shape)))
# noinspection PyArgumentList
correlator_symmetries = (HashableArray(geneqs.utils.indexing.get_cubeperms_cubical3d(shape, 2)),
HashableArray(geneqs.utils.indexing.get_bondperms_cubical3d(shape, 2)))
# noinspection PyArgumentList
loops = (HashableArray(geneqs.utils.indexing.get_strings_cubical3d(0, shape)),
HashableArray(geneqs.utils.indexing.get_strings_cubical3d(1, shape)),
HashableArray(geneqs.utils.indexing.get_strings_cubical3d(2, shape)))
# noinspection PyArgumentList
loop_symmetries = (HashableArray(geneqs.utils.indexing.get_xstring_perms3d(shape, 2)),
HashableArray(geneqs.utils.indexing.get_ystring_perms3d(shape, 2)),
HashableArray(geneqs.utils.indexing.get_zstring_perms3d(shape, 2)))
alpha = 1 / 4
default_kernel_init = jax.nn.initializers.normal(0.01)
cRBM = geneqs.models.CheckerLoopCRBM_2(symmetries=perms,
correlators=correlators,
correlator_symmetries=correlator_symmetries,
loops=loops,
loop_symmetries=loop_symmetries,
alpha=alpha,
kernel_init=default_kernel_init,
bias_init=default_kernel_init,
param_dtype=complex)
model = cRBM
# create custom update rule
single_rule = nk.sampler.rules.LocalRule()
cube_rule = geneqs.sampling.update_rules.MultiRule(geneqs.utils.indexing.get_cubes_cubical3d(shape, shift=2))
xstring_rule = geneqs.sampling.update_rules.MultiRule(geneqs.utils.indexing.get_strings_cubical3d(0, shape))
ystring_rule = geneqs.sampling.update_rules.MultiRule(geneqs.utils.indexing.get_strings_cubical3d(1, shape))
zstring_rule = geneqs.sampling.update_rules.MultiRule(geneqs.utils.indexing.get_strings_cubical3d(2, shape))
# noinspection PyArgumentList
weighted_rule = geneqs.sampling.update_rules.WeightedRule((0.51, 0.25, 0.08, 0.08, 0.08),
[single_rule,
cube_rule,
xstring_rule,
ystring_rule,
zstring_rule])
# %% for every h, create model, load params, calculate observables
observables = geneqs.utils.eval_obs.ObservableCollector(key_names=("hx", "hy", "hz"))
for h in tqdm(field_strengths, "external_field"):
h = tuple(h)
print(f"evaluation for field={h}")
checkerboard = geneqs.operators.checkerboard.Checkerboard(hilbert, shape, h)
sampler = nk.sampler.MetropolisSampler(hilbert, rule=weighted_rule, n_chains=n_chains, dtype=jnp.int8)
vqs = nk.vqs.MCState(sampler, model, n_samples=n_samples, n_discard_per_chain=n_discard_per_chain)
sampler_state = vqs.sampler_state
with open(f"{save_dir}/vqs_{eval_model}_L{shape}_h{h}.mpack", 'rb') as file:
vqs = flax.serialization.from_bytes(vqs, file.read())
rank_sigmas = vqs.sampler_state.σ.reshape(n_ranks, -1, vqs.sampler_state.σ.shape[-1])
rank_sampler_state = vqs.sampler_state.replace(σ=rank_sigmas[rank])
vqs.sampler_state = rank_sampler_state
vqs.n_chains_per_rank = vqs.sampler_state.σ.shape[0]
vqs.chunk_size = chunk_size
vqs.n_samples = n_samples
print(vqs.n_samples, vqs.n_samples_per_rank, vqs.sampler.n_chains, vqs.sampler.n_chains_per_rank, vqs.sampler_state.σ.shape)
# calculate energy and variance of energy
energy_nk = geneqs.utils.eval_obs.expect_batched(vqs, checkerboard, n_samples=n_samples, chunk_size=chunk_size)
observables.add_nk_obs("energy", h, energy_nk)
# calculate magnetization
magnetization_nk = geneqs.utils.eval_obs.expect_batched(vqs, magnetization, n_samples=n_samples, chunk_size=chunk_size)
observables.add_nk_obs("mag", h, magnetization_nk)
# calculate cube parts of the hamiltonian
xcubes_nk = geneqs.utils.eval_obs.expect_batched(vqs, xcubes, n_samples=n_samples, chunk_size=chunk_size)
observables.add_nk_obs("xcubes", h, xcubes_nk)
zcubes_nk = geneqs.utils.eval_obs.expect_batched(vqs, zcubes, n_samples=n_samples, chunk_size=chunk_size)
observables.add_nk_obs("zcubes", h, zcubes_nk)
print(f"donw with {h}")
if rank == 0:
save_array = observables.obs_to_array(separate_keys=False)[-1].reshape(1, -1)
with open(f"{save_dir}/L{shape}_{eval_model}_eval_obs.txt", "ab") as f:
if os.path.getsize(f"{save_dir}/L{shape}_{eval_model}_eval_obs.txt") == 0:
np.savetxt(f, save_array, header=" ".join(observables.key_names + observables.obs_names),
comments="")
else:
np.savetxt(f, save_array)
# calculate absolute magnetization
# abs_magnetization_nk = geneqs.utils.eval_obs.expect_batched(vqs, abs_magnetization, n_samples=n_samples, chunk_size=chunk_size)
# observables.add_nk_obs("abs_mag", h, abs_magnetization_nk)
# gather local estimators as each rank calculates them based on their own samples_per_rank
# if np.any((h == hist_fields).all(axis=1)):
#
# energy_locests = []
# mag_locests = []
# vqs.n_samples = 2*chunk_size
#
# for _ in tqdm(range(1)):
# vqs.sample()
# energy_locests.append(np.array(comm.gather(vqs.local_estimators(checkerboard), root=0)).flatten().real)
# mag_locests.append(np.array(comm.gather(vqs.local_estimators(magnetization), root=0)).flatten().real)
# energy_locests = np.concatenate(energy_locests)
#
#
# observables.add_hist("epsite", h, np.histogram(np.asarray(energy_locests) / hilbert.size, n_bins))
# observables.add_hist("mag", h, np.histogram(np.asarray(mag_locests), n_bins))
#
# # %% save histograms
# if rank == 0:
# print(energy_locests.shape)
# for hist_name, _ in observables.histograms.items():
# np.save(f"{save_dir}/hist_{hist_name}_L{shape}_{eval_model}_{label}.npy",
# observables.hist_to_array(hist_name))
# if rank == 0:
# save_array = observables.obs_to_array(separate_keys=False)
# np.savetxt(f"{save_dir}/L{shape}_{eval_model}_eval_obs.txt", save_array,
# header=" ".join(observables.key_names + observables.obs_names), comments="")