-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_inference_sgd.py
executable file
·160 lines (130 loc) · 5.14 KB
/
run_inference_sgd.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
from src.model import configSimulation, simulationLoopUnsafe
from numpyro.infer.reparam import TransformReparam
import os
import sys
import time
from functools import partial
from jax import jit, grad, jacfwd
import numpyro.distributions as dist
import jax.numpy as jnp
import jax
import numpy as np
import itertools
import optax
from flax.training.train_state import TrainState
os.chdir(os.path.dirname(__file__))
jax.config.update("jax_enable_x64", True)
config_filename = ""
if len(sys.argv) == 1:
# base cases
#modelname = "single-artery"
#modelname = "tapering"
#modelname = "conjunction"
#modelname = "bifurcation"
#modelname = "aspirator"
# openBF-hub
modelname = "test/adan56/adan56.yml"
# vascularmodels.com
#modelname = "0007_H_AO_H"
#modelname = "0029_H_ABAO_H"
#modelname = "0053_H_CERE_H"
input_filename = "test/" + modelname + "/" + modelname + ".yml"
else:
config_filename = "test/" + sys.argv[1] + "/" + sys.argv[1] + ".yml"
verbose = True
(N, B, J,
sim_dat, sim_dat_aux,
sim_dat_const, sim_dat_const_aux,
timepoints, conv_tol, Ccfl, edges, input_data, rho,
masks, strides, edges,
vessel_names, cardiac_T) = configSimulation(config_filename, verbose)
if verbose:
starting_time = time.time_ns()
sim_loop_old_jit = partial(jit, static_argnums=(0, 1, 12))(simulationLoopUnsafe)
sim_dat_obs, t_obs, P_obs = sim_loop_old_jit(N, B,
sim_dat, sim_dat_aux,
sim_dat_const, sim_dat_const_aux,
Ccfl, input_data, rho,
masks, strides, edges,
upper=120000)
Ccfl = 0.5
R_index = 1
var_index = 7
R1 = sim_dat_const[var_index,strides[R_index,1]]
#R_scales = np.linspace(1.1*R1, 2*R1, 16)
R_scales = np.linspace(0.1, 10, 8)
def simLoopWrapper(params):
R = params[0]*R1
ones = jnp.ones(strides[R_index,1]-strides[R_index,0]+4)
sim_dat_const_new = jnp.array(sim_dat_const)
sim_dat_const_new = sim_dat_const_new.at[var_index,strides[R_index,0]-2:strides[R_index,1]+2].set(R*ones)
_, _, P = sim_loop_old_jit(N, B,
sim_dat, sim_dat_aux,
sim_dat_const_new, sim_dat_const_aux,
Ccfl, input_data, rho,
masks, strides, edges,
upper=120000)
#jax.debug.print("P={x}", x=P)
#jax.debug.print("R={x}", x=R)
#jax.debug.print("P_obs={x}", x=P_obs)
return P
results_folder = "results/inference_ensemble_det"
if not os.path.isdir(results_folder):
os.makedirs(results_folder, mode = 0o777)
network_properties = {
#"tx": [optax.adam, optax.sgd, optax.lars],
#"tx": [optax.adabelief, optax.adadelta, optax.adafactor, optax.adagrad,
# optax.adamw, optax.adamax, optax.adamaxw, optax.amsgrad],
#"tx": [optax.adagrad,
# optax.adamw, optax.adamax, optax.adamaxw, optax.amsgrad],
"tx": [optax.adabelief],
"learning_rate": [1e3, 1e4, 1e5, 1e6, 1e7, 1e8],
"epochs": [2000]
}
settings = list(itertools.product(*network_properties.values()))
results_folder = "results/inference_ensemble_sgd"
if not os.path.isdir(results_folder):
os.makedirs(results_folder, mode = 0o777)
for set_num, setup in enumerate(settings):
print("###################################", set_num, "###################################")
results_file = results_folder + "/setup_" + str(setup[0].__name__) + "_" + str(setup[1]) + "_" + str(setup[2]) +"_test.txt"
model = simLoopWrapper
variables = [R_scales[int(sys.argv[2])]]
tx = setup[0]
y = P_obs
x = simLoopWrapper
state = TrainState.create(
apply_fn=model,
params=variables,
tx=tx(setup[1]))
def loss_fn(params, x, y):
predictions = state.apply_fn(params)
loss = jnp.log(optax.l2_loss(predictions=predictions, targets=y).mean()/optax.l2_loss(predictions=np.ones_like(y), targets=y).mean()+1)
return loss
for _ in range(setup[2]):
grads = jax.jacfwd(loss_fn)(state.params, x, y)
#print(grads)
state = state.apply_gradients(grads=grads)
#print(state)
print(loss_fn(state.params, x, y))
file = open(results_file, "a")
file.write(str(R_scales[int(sys.argv[2])]) + " " + str(state.params[0]) + " " + str(R1) + "\n")
file.close()
#for (j, learning_rate) in enumerate(learning_rates):
# R_star = R_scales[int(sys.argv[2])]
# for i in range(1000):
# print(j,i)
# gradient = sim_loop_wrapper_jit(R_star)
# R_star -= learning_rate*gradient
#
# results_file = results_folder + "/setup_" + str(learning_rate) + ".txt"
# file = open(results_file, "a")
# file.write(str(R_scales[int(sys.argv[2])]) + " " + str(R_star) + " " + str(R1) + "\n")
# file.close()
network_properties = {
"sigma": [1e-5],
"scale": [10],
"num_warmup": np.arange(10, 110, 10),
"num_samples": np.arange(100, 1100, 100),
"num_chains": [1]
}