-
Notifications
You must be signed in to change notification settings - Fork 1
/
hdbayesopt.py
158 lines (139 loc) · 6.88 KB
/
hdbayesopt.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
import os.path as osp
import pickle
import sys
import time
import matplotlib.pyplot as plt
import numpy as np
import scipy.io
import torch
from skimage.filters import threshold_otsu
sys.path.append('/bayesopt/')
from bayes_opt import BayesianOptimization
import cardiacmodel
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
def hdbayesopt(vae, dpath, rpath, files, p_dim, acq_list=None, niter=None,
inipts=None, z_mu_1=None, z_var_1=None, verbose=0):
"""Run High Dimensional Bayesion Optimization Experiments
Args:
vae: trained model
dpath: input path fro experiments
rpath: output path fore optimization results
files: files of the experiments to run
p_dim: number of meshfree node in given heart
acq_list: acquisation functions to use
niter: number of iterations for optimization
inipts: number of initial input points
"""
latent_dim = vae.latent_dim
# set the total number of initial points
# and iterations for optimization based on the size of latent_dim
map_dim_niter = {2: 85, 3: 150, 5: 300}
map_dim_inipts = {2: 5, 3: 10, 5: 20}
if not niter:
niter = map_dim_niter[latent_dim]
if not inipts:
niter = map_dim_inipts[latent_dim]
if not acq_list:
acq_list = ('ei')
acq_list = tuple(acq_list)
num_acq = len(acq_list) # number of acquisation function
num_exps = len(files) # number of experiments
parUnknownId = list(range(1, latent_dim + 1)) # if for each unknown
bounds = [(-4, 4) for ij in parUnknownId] # bounds on the optimization
parUnknownId = [str(ij) for ij in parUnknownId] # convert id to string
# initialize variables to collect data
timetaken = np.zeros((num_exps, num_acq))
dicecoeff = np.zeros((num_exps, num_acq))
rmse = np.zeros((num_exps, num_acq))
fopt = np.zeros((num_exps, num_acq)) # optimum value of the function
paramEstRes = np.zeros((num_exps, num_acq, p_dim)) # estimated parameter at original dim
paramEstRes_z = np.zeros((num_exps, num_acq, latent_dim)) # estimated z
paramGT = np.zeros((num_exps, num_acq, p_dim)) # ground truth
# loop through each experiment for parameter estimation
for i in range(num_exps):
# read the experiment that is in matlab format
fname = files[i] + '.mat'
matFiles = scipy.io.loadmat(dpath + '/' + fname, squeeze_me=True, struct_as_record=False)
parTrue = matFiles['parTrue']
obs = matFiles['obs']
simu = matFiles['simu']
corMfree = matFiles['corMfree']
# instance of cardaic model
cardiac_model = cardiacmodel.CardiacModel(simu, obs, parTrue, corMfree, maskidx_12lead=0,
device=device)
thresh_gt = 0.18 # threshold_otsu(cardiac_model.parTrue)
idx_gt = np.where(cardiac_model.true_par >= thresh_gt)[0]
# t,tmp,bsp=cardiac_model.simulate_ecg(parTrue)
# for each exp loop through each acq func
for j in range(num_acq):
gp_surr = BayesianOptimization(cardiac_model.compute_objfunc,
dict(zip(parUnknownId, bounds)), vae, verbose=verbose)
acq_func = acq_list[j]
tstart = time.time()
if (acq_func == 'ei'):
gp_surr.maximize(init_points=inipts, n_iter=niter, acq='ei', xi=0.0001)
elif (acq_func == 'ei_prior'):
gp_surr.maximize(init_points=inipts, n_iter=niter, acq='ei_prior', xi=1.0)
elif (acq_func == 'ei_post_agg'):
gp_surr.maximize(init_points=inipts, n_iter=niter, acq='ei_post_agg',
z_m=z_mu_1, z_v=z_var_1, xi=1.0)
else:
# TODO bring implementation of other acquisation function in thsi code
print('incorrect acq func')
tend = time.time()
# optimum found
xmax = gp_surr.res['max']['max_params']
z_mu = torch.from_numpy(np.array([xmax]*vae.batch_size)).float()
z_mu = z_mu.to(device)
with torch.no_grad():
x_mean = vae.decode(z_mu)
x_mean = (x_mean[0]).cpu().numpy()
# if use_cpd:
# x_mean = x_mean[correspondance]
# compute dice coefficient and rmse
thresh_c = threshold_otsu(x_mean)
idx_c = np.where(x_mean >= thresh_c)[0]
rmse_temp = np.sqrt(((cardiac_model.true_par - x_mean) ** 2).mean())
dicecoeff_temp = 2 * len(np.intersect1d(idx_gt, idx_c)) / (len(idx_gt) + len(idx_c))
rmse[i, j] = rmse_temp
dicecoeff[i, j] = dicecoeff_temp
paramEstRes[i, j, :] = x_mean
paramEstRes_z[i, j, :] = xmax
fopt[i, j] = gp_surr.res['max']['max_val']
paramGT[i, j, :] = cardiac_model.true_par
timetaken[i, j] = tend - tstart
# plot of the true and estimated paraemters
fname_save = osp.join(rpath, files[i] + '_' + acq_func + '.png')
fig = plt.figure()
ax1 = fig.add_subplot(1, 2, 1, projection='3d')
cardiac_model.plot_gt(ax1)
ax2 = fig.add_subplot(1, 2, 2, projection='3d')
cardiac_model.plot_param(x_mean, ax=ax2)
fig.savefig(fname_save, dpi=300, bbox_inches='tight', transparent=True)
print('exp #{:03d} with {}, dc:{:.4f}, rmse:{:.4f}, time:{:.4f}'.format(i + 1, acq_func,
dicecoeff_temp, rmse_temp,
timetaken[i, j] / 60))
# print('exp # {}' + str(i+1)+ ': ' + acq_func + ': '
# + str(dicecoeff_temp) + ', ' str(rmse_temp) + ', ' + str(tend-tstart))
del gp_surr
del xmax
del z_mu
del x_mean
del thresh_c
del idx_c
del rmse_temp
del dicecoeff_temp
# save optimization result in pickle format
fname_save = osp.join(rpath, files[i] + '_' + str(latent_dim) + '_d.pkl')
with open(fname_save, 'wb') as output:
pickle.dump(fname, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(cardiac_model, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(paramEstRes, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(paramEstRes_z, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(rmse, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(dicecoeff, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(fopt, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(paramGT, output, pickle.HIGHEST_PROTOCOL)
pickle.dump(timetaken, output, pickle.HIGHEST_PROTOCOL)
del cardiac_model
del idx_gt