-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathplot_dro_baseline.py
216 lines (172 loc) · 6.96 KB
/
plot_dro_baseline.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
205
206
207
208
209
210
211
212
213
214
import pickle
import yaml
import torch
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from collections import defaultdict
import seaborn as sns
import itertools
import torch
import os
import argparse
evd, unc, evd_sm = True, True, True
fill = True
parser = argparse.ArgumentParser(
prog = 'MEVDRONet',
description = 'Computes adversarial risk for some data.')
parser.add_argument('filename')
args = parser.parse_args()
with open(args.filename, 'r') as f:
params = yaml.safe_load(f)
d = params['d']
n_epochs = params['n_epochs']
n_lam = params['n_lam']
width = params['width']
n_data = params['n_data']
block_size = params['block_size']
if block_size > 0:
n_data = n_data // block_size
if block_size == 100:
n_data = 20
n_max = params['n_max']
rate = params['rate']
use_softmax = params['use_softmax']
experiment = params['experiment']
risk = params['risk']
n_eps = params['n_eps']
n_runs = params['n_runs']
eps_max = params['eps_max']
data_file = params['data_file']
data_file_short_period = params["data_file_short_period"]
cost_norm = params['cost_norm']
try:
eps_coef = params['eps_coef']
except:
eps_coef = None
synthetic_rate = params['synthetic_rate'] #0
file = open(data_file_short_period + ".p", 'rb')
weekly_base_data = pickle.load(file)
file.close()
weekly_base_data_df = pd.DataFrame(weekly_base_data)
weekly_base_data_df.plot(figsize=(24,8), alpha = 0.5)
plt.savefig("weekly_max_return_data.png")
file = open(data_file + ".p", 'rb')
yearly_base_data = pickle.load(file)
file.close()
yearly_base_data_df = pd.DataFrame(yearly_base_data)
yearly_base_data_df.plot(figsize=(24,8), alpha = 0.5)
plt.savefig("yearly_max_return_data.png")
n_data = weekly_base_data.shape[0]
true_key = 'true_risk'
pop_key = 'p0_risk'
adv_key = 'adv_risk'
evd_file = data_file + '_evd_{}_gen_eps{}_data{}_blocksize{}/'.format(risk, eps_max, n_max, block_size)
evd_sm_file = data_file + '_evd-sm_{}_gen_eps{}_data{}_blocksize{}/'.format(risk, eps_max, n_max, block_size)
unc_file = data_file + '_unc_{}_gen_eps{}_data{}_blocksize{}/'.format(risk, eps_max, n_max, block_size)
####LOAD EVD DATA#######
try:
print(evd_file + 'stats{}_{}.p'.format(eps_max, n_data))
evd_load = open(evd_file + 'stats{}_{}.p'.format(eps_max, n_data), 'rb')
data = pickle.load(evd_load)
evd_load.close()
true_evd = data[true_key]
pop_evd = data[pop_key]
adv = data[adv_key]
adv = adv * (adv < 1e3)
x = np.stack(data['losses'])[:,0]
x_domain_evd = (x / true_evd.mean(-1))
except :
adv, evd = None, False
print('EVD stats not found, skipping')
#########################
# ####LOAD EVD-SM DATA#######
try:
print(evd_sm_file + 'stats{}_{}.p'.format(eps_max, n_data))
evd_sm_load = open(evd_sm_file + 'stats{}_{}.p'.format(eps_max, n_data), 'rb')
datasm = pickle.load(evd_sm_load)
evd_sm_load.close()
true_sm = datasm[true_key]
pop_sm = datasm[pop_key]
advsm = datasm[adv_key]
advsm = advsm * (advsm < 1e3)
x_sm = np.stack(datasm['losses'])[:,0]
x_domain_sm = (x_sm / true_sm.mean(-1))
x_domain_max = np.around(torch.max(x_sm / true_sm.mean(-1)).detach().numpy())
x_ticks = np.arange(x_domain_max + 1, dtype=int)
except :
advsm, evd_sm = None, False
print('EVD-sm stats not found, skipping')
# #########################
# ####LOAD UNC DATA#######
try:
print(unc_file + 'stats{}_{}.p'.format(eps_max, n_data))
unc_load = open(unc_file + 'stats{}_{}.p'.format(eps_max, n_data), 'rb')
datau = pickle.load(unc_load)
unc_load.close()
true_unc = datau[true_key]
pop_unc = datau[pop_key]
advu = datau[adv_key]
advu = advu * (advu.abs() < 1e3)
x_unc = np.stack(datau['losses'])[:,0]
x_domain_unc = (x_unc / true_unc.mean(-1))
x_domain_max = np.around(torch.max(x_unc / true_unc.mean(-1)).detach().numpy())
x_ticks = np.arange(x_domain_max + 1, dtype=int)
except :
advu, unc = None, False
print('Unconstrained stats not found, skipping')
# #########################
start_domain = 0
fig, ax = plt.subplots(figsize=(10,6))
error_P0 = torch.abs(pop_evd - true_evd)
error_P0_lower = np.nanmean(np.log(error_P0), -1) - np.nanstd(np.log(error_P0), -1)
error_P0_upper = np.nanmean(np.log(error_P0), -1) + np.nanstd(np.log(error_P0), -1)
ax.plot(x_domain_evd, np.nanmean(np.log(error_P0), -1), color = "red", marker ='x', label=r'$P_0$: Non-DRO EVD Risk')
if fill:
ax.fill_between(x_domain_evd, error_P0_lower, error_P0_upper, color = "red", alpha=0.3)
#####PLOT EVD#######
if evd:
error_Pstar = adv - true_evd
error_Pstar[0] = error_P0[0]
error_Pstar_lower = np.nanmean(np.log(error_Pstar), -1) - np.nanstd(np.log(error_Pstar), -1)
error_Pstar_upper = np.nanmean(np.log(error_Pstar), -1) + np.nanstd(np.log(error_Pstar), -1)
ax.plot(x_domain_evd, np.log(np.nanmean(error_Pstar, -1)), color = "blue", marker ='*', label=r'$P_\star$: DRO EVD Risk')
if fill:
ax.fill_between(x_domain_evd, error_Pstar_lower, error_Pstar_upper, color = "blue", alpha=0.2)
#####################
#####PLOT EVD-SM#####
if evd_sm:
error_PstarUni = advsm - true_sm
error_PstarUni[0] = error_P0[0]
error_PstarUni_lower = np.nanmean(np.log(error_PstarUni), -1) - np.nanstd(np.log(error_PstarUni), -1)
error_PstarUni_upper = np.nanmean(np.log(error_PstarUni), -1) + np.nanstd(np.log(error_PstarUni), -1)
ax.plot(x_domain_sm, np.log(np.nanmean(error_PstarUni, -1)), color = "orange", marker ='*', label='$P_\star (\mathbb{E}[w_i] \approx \frac{1}{d})$: DRO EVD\nRisk w/ Uniform Margins')
if fill:
ax.fill_between(x_domain_sm, error_PstarUni_lower, error_PstarUni_upper, color = "orange", alpha=0.2)
#####################
#####PLOT UNC#####
if unc:
error_Punc = advu - true_unc
error_Punc[0] = error_P0[0]
error_Punc_lower = np.nanmean(np.log(error_Punc), -1) - np.nanstd(np.log(error_Punc), -1)
error_Punc_upper = np.nanmean(np.log(error_Punc), -1) + np.nanstd(np.log(error_Punc), -1)
ax.plot(x_domain_unc, np.log(np.nanmean(error_Punc, -1)), color = "green", marker ='*', label=r'$P_\star (unc)$: Non-MEV Risk')
if fill:
ax.fill_between(x_domain_unc, error_Punc_lower, error_Punc_upper, color = "green", alpha=0.1)
#####################
y_domain_max = np.around(np.max(np.log(np.nanmean(error_Pstar, -1))))
print(y_domain_max)
y_ticks = np.arange(y_domain_max + 1, dtype=int)
ax.spines['top'].set_color('none')
ax.spines['right'].set_position(('axes', 1.0))
ax.spines['left'].set_color('none')
ax.spines['bottom'].set_position(('axes', 0.0))
plt.xticks(x_ticks)
ax.yaxis.tick_right()
plt.xlabel(r'$\delta$ Normalized by True Risk', fontsize = 18)
ax.yaxis.set_label_position("right")
plt.ylabel(r'$Log \vert \mathbb{E}[\ell(X_{P_{true}})] - \mathbb{E}[\ell(X_{P_{model}})] \vert$', fontsize = 18)
plt.title("Daily Returns Expected Risk Evaluated Over\nIncreasing Uncertainty ($\delta$)", fontsize = 20)
plt.legend(loc= "lower right", fontsize = 13)
plt.tight_layout()
plt.savefig("baseline.png")