-
Notifications
You must be signed in to change notification settings - Fork 20
/
Copy pathexperimental_utils.py
175 lines (156 loc) · 9.38 KB
/
experimental_utils.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
import os
import time
import numpy as np
from utils import eval_causal_structure, eval_causal_structure_binary
from datetime import date
from training import training_procedure_trgc
from plotting_utils import visualise_gen_coeffs_lotka_volterra, visualise_gen_coeffs_linear_var
def run_grid_search(lambdas: np.ndarray, gammas: np.ndarray, datasets: list, structures: list, K: int,
num_hidden_layers: int, hidden_layer_size: int, num_epochs: int, batch_size: int,
initial_lr: float, beta_1: float, beta_2: float, seed: int, signed_structures=None):
"""
Evaluates GVAR model across a range of hyperparameters.
@param lambdas: values for the sparsity-inducing penalty parameter.
@param gammas: values for the smoothing penalty parameter.
@param datasets: list of time series datasets.
@param structures: ground truth GC structures.
@param K: model order.
@param num_hidden_layers: number of hidden layers.
@param hidden_layer_size: number of units in a hidden layer.
@param num_epochs: number of training epochs.
@param batch_size: batch size.
@param initial_lr: learning rate.
@param seed: random generator seed.
@param signed_structures: ground truth signs of GC interactions.
"""
# Logging
logdir = "logs/" + str(date.today()) + "_" + str(round(time.time())) + "_validation_gvar"
print("Log directory: " + logdir + "/")
os.mkdir(path=logdir)
np.savetxt(fname=logdir + "/lambdas.csv", X=lambdas)
np.savetxt(fname=logdir + "/gammas.csv", X=gammas)
# For binary structures
mean_accs = np.zeros((len(lambdas), len(gammas)))
sd_accs = np.zeros((len(lambdas), len(gammas)))
mean_bal_accs = np.zeros((len(lambdas), len(gammas)))
sd_bal_accs = np.zeros((len(lambdas), len(gammas)))
mean_precs = np.zeros((len(lambdas), len(gammas)))
sd_precs = np.zeros((len(lambdas), len(gammas)))
mean_recs = np.zeros((len(lambdas), len(gammas)))
sd_recs = np.zeros((len(lambdas), len(gammas)))
# For continuous structures
mean_aurocs = np.zeros((len(lambdas), len(gammas)))
sd_aurocs = np.zeros((len(lambdas), len(gammas)))
mean_auprcs = np.zeros((len(lambdas), len(gammas)))
sd_auprcs = np.zeros((len(lambdas), len(gammas)))
# For effect signs
if signed_structures is not None:
mean_bal_accs_pos = np.zeros((len(lambdas), len(gammas)))
sd_bal_accs_pos = np.zeros((len(lambdas), len(gammas)))
mean_bal_accs_neg = np.zeros((len(lambdas), len(gammas)))
sd_bal_accs_neg = np.zeros((len(lambdas), len(gammas)))
n_datasets = len(datasets)
print("Iterating through " + str(len(lambdas)) + " x " + str(len(gammas)) + " grid of parameters...")
for i in range(len(lambdas)):
lmbd_i = lambdas[i]
for j in range(len(gammas)):
gamma_j = gammas[j]
print("λ = " + str(lambdas[i]) + "; γ = " + str(gammas[j]) + "; " +
str((i * len(gammas) + j) / (len(gammas) * len(lambdas)) * 100) + "% done")
accs_ij = []
bal_accs_ij = []
prec_ij = []
rec_ij = []
aurocs_ij = []
auprcs_ij = []
if signed_structures is not None:
bal_accs_pos_ij = []
bal_accs_neg_ij = []
for l in range(n_datasets):
d_l = datasets[l]
a_l = structures[l]
if signed_structures is None:
a_hat_l, a_hat_l_, coeffs_full_l = training_procedure_trgc(data=d_l, order=K,
hidden_layer_size=hidden_layer_size,
end_epoch=num_epochs, lmbd=lmbd_i,
gamma=gamma_j, batch_size=batch_size,
seed=(seed + i + j),
num_hidden_layers=num_hidden_layers,
initial_learning_rate=initial_lr,
beta_1=beta_1, beta_2=beta_2,
verbose=False)
else:
a_l_signed = signed_structures[l]
a_hat_l, a_hat_l_, a_hat_l_signed, coeffs_full_l = training_procedure_trgc(data=d_l, order=K,
hidden_layer_size=hidden_layer_size,
end_epoch=num_epochs, lmbd=lmbd_i,
gamma=gamma_j, batch_size=batch_size,
seed=(seed + i + j),
num_hidden_layers=num_hidden_layers,
initial_learning_rate=initial_lr,
beta_1=beta_1, beta_2=beta_2,
verbose=False, signed=True)
acc_l, bal_acc_l, prec_l, rec_l = eval_causal_structure_binary(a_true=a_l, a_pred=a_hat_l)
auroc_l, auprc_l = eval_causal_structure(a_true=a_l, a_pred=a_hat_l_)
accs_ij.append(acc_l)
bal_accs_ij.append(bal_acc_l)
prec_ij.append(prec_l)
rec_ij.append(rec_l)
aurocs_ij.append(auroc_l)
auprcs_ij.append(auprc_l)
print("Dataset #" + str(l + 1) + "; Acc.: " + str(np.round(acc_l, 4)) + "; Bal. Acc.: " +
str(np.round(bal_acc_l, 4)) + "; Prec.: " + str(np.round(prec_l, 4)) + "; Rec.: " +
str(np.round(rec_l, 4)) + "; AUROC: " + str(np.round(auroc_l, 4)) + "; AUPRC: " +
str(np.round(auprc_l, 4)), end='\r')
if signed_structures is not None:
if len(a_hat_l_signed.shape) == 3:
a_hat_l_signed = np.mean(a_hat_l_signed, axis=0)
_, bal_acc_pos, __, ___ = eval_causal_structure_binary(a_true=(a_l_signed > 0) * 1.0,
a_pred=(a_hat_l_signed > 0) * 1.0)
_, bal_acc_neg, __, ___ = eval_causal_structure_binary(a_true=(a_l_signed < 0) * 1.0,
a_pred=(a_hat_l_signed < 0) * 1.0)
bal_accs_pos_ij.append(bal_acc_pos)
bal_accs_neg_ij.append(bal_acc_neg)
print()
mean_accs[i, j] = np.mean(accs_ij)
print("Acc. :" + str(mean_accs[i, j]))
sd_accs[i, j] = np.std(accs_ij)
mean_bal_accs[i, j] = np.mean(bal_accs_ij)
print("Bal. Acc. :" + str(mean_bal_accs[i, j]))
sd_bal_accs[i, j] = np.std(bal_accs_ij)
mean_precs[i, j] = np.mean(prec_ij)
print("Prec. :" + str(mean_precs[i, j]))
sd_precs[i, j] = np.std(prec_ij)
mean_recs[i, j] = np.mean(rec_ij)
print("Rec. :" + str(mean_recs[i, j]))
sd_recs[i, j] = np.std(rec_ij)
mean_aurocs[i, j] = np.mean(aurocs_ij)
print("AUROC :" + str(mean_aurocs[i, j]))
sd_aurocs[i, j] = np.std(aurocs_ij)
mean_auprcs[i, j] = np.mean(auprcs_ij)
print("AUPRC :" + str(mean_auprcs[i, j]))
sd_auprcs[i, j] = np.std(auprcs_ij)
if signed_structures is not None:
mean_bal_accs_pos[i, j] = np.mean(bal_accs_pos_ij)
print("BA (pos.) :" + str(mean_bal_accs_pos[i, j]))
sd_bal_accs_pos[i, j] = np.std(bal_accs_pos_ij)
mean_bal_accs_neg[i, j] = np.mean(bal_accs_neg_ij)
print("BA (neg.) :" + str(mean_bal_accs_neg[i, j]))
sd_bal_accs_neg[i, j] = np.std(bal_accs_neg_ij)
np.savetxt(fname=logdir + "/mean_accs.csv", X=mean_accs)
np.savetxt(fname=logdir + "/sd_accs.csv", X=sd_accs)
np.savetxt(fname=logdir + "/mean_bal_accs.csv", X=mean_bal_accs)
np.savetxt(fname=logdir + "/sd_bal_accs.csv", X=sd_bal_accs)
np.savetxt(fname=logdir + "/mean_precs.csv", X=mean_precs)
np.savetxt(fname=logdir + "/sd_precs.csv", X=sd_precs)
np.savetxt(fname=logdir + "/mean_recs.csv", X=mean_recs)
np.savetxt(fname=logdir + "/sd_recs.csv", X=sd_recs)
np.savetxt(fname=logdir + "/mean_aurocs.csv", X=mean_aurocs)
np.savetxt(fname=logdir + "/sd_aurocs.csv", X=sd_aurocs)
np.savetxt(fname=logdir + "/mean_auprcs.csv", X=mean_auprcs)
np.savetxt(fname=logdir + "/sd_auprcs.csv", X=sd_auprcs)
if signed_structures is not None:
np.savetxt(fname=logdir + "/mean_bal_accs_pos.csv", X=mean_bal_accs_pos)
np.savetxt(fname=logdir + "/sd_bal_accs_pos.csv", X=sd_bal_accs_pos)
np.savetxt(fname=logdir + "/mean_bal_accs_neg.csv", X=mean_bal_accs_neg)
np.savetxt(fname=logdir + "/sd_bal_accs_neg.csv", X=sd_bal_accs_neg)