-
Notifications
You must be signed in to change notification settings - Fork 0
/
analysis.py
254 lines (208 loc) · 11.9 KB
/
analysis.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
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
import pandas as pd
import numpy as np
import sys
import os
import matplotlib.pyplot as plt
import seaborn as sns
import pickle
import argparse
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Analysis of the results')
parser.add_argument('--file_type', type=str, default='pdf', help='File type of the figures')
args = parser.parse_args()
file_type = args.file_type
# ================================== EXP 1 ===================================
# Convergence analysis
ffo_result, cvxpylayer_result = {}, {}
ffo_result_mean, cvxpylayer_result_mean = {}, {}
ffo_result_std, cvxpylayer_result_std = {}, {}
grad_differences = {}
eps = 0.01
ydim_list = [5, 10, 20, 50, 100, 200, 500] # , 800, 1000] # list(range(100,1000,100))
directory_path = 'exp1/'
# directory_path = 'exp1_bilinear/'
seed_list = list(set(range(1,11,1))) # - set([2,9,29]))
for ydim in ydim_list:
directory_name = directory_path + 'ydim{}'.format(ydim)
# directory_name_exp3 = 'exp3_bilinear/' + 'ydim{}'.format(ydim)
# Initialize the dictionary
ffo_result[ydim], cvxpylayer_result[ydim] = [], []
grad_differences[ydim] = []
# Read the results
ffo_filename_list = ['ffo_eps{}_seed{}.txt'.format(eps, seed) for seed in seed_list]
ffo_grad_filename_list = ['ffo_eps{}_seed{}.pickle'.format(eps, seed) for seed in seed_list]
cvxpylayer_filename_list = ['cvxpylayer_eps{}_seed{}.txt'.format(eps, seed) for seed in seed_list]
cvxpylayer_gradient_filename_list = ['cvxpylayer_eps{}_seed{}.pickle'.format(eps, seed) for seed in seed_list]
for filename in ffo_filename_list:
df = pd.read_csv('results/' + directory_name + '/' + filename, header=None, names=['iteration', 'loss', 'time', 'time1', 'time2'], skiprows=1)
ffo_result[ydim].append(df.values[:,:3] - df.values[:,:3].min(axis=0))
for filename in cvxpylayer_filename_list:
df = pd.read_csv('results/' + directory_name + '/' + filename)
cvxpylayer_result[ydim].append(df.values[:,:3] - df.values[:,:3].min(axis=0))
ffo_result_mean[ydim] = np.mean(ffo_result[ydim], axis=0)
ffo_result_std[ydim] = np.std(ffo_result[ydim], axis=0)
cvxpylayer_result_mean[ydim] = np.mean(cvxpylayer_result[ydim], axis=0)
cvxpylayer_result_std[ydim] = np.std(cvxpylayer_result[ydim], axis=0)
# Compute gradient difference
for filename in ffo_grad_filename_list:
ffo_grad_list = pickle.load(open('results/' + directory_name + '/' + filename, 'rb'))
for filename in cvxpylayer_gradient_filename_list:
cvxpylaer_grad_list = pickle.load(open('results/' + directory_name + '/' + filename, 'rb'))
for ffo_grad, cvxpylayer_grad in zip(ffo_grad_list, cvxpylaer_grad_list):
grad_differences[ydim].append(np.linalg.norm(ffo_grad - cvxpylayer_grad))
# grad_differences[ydim] = np.convolve(grad_differences[ydim], np.ones(10)/10, mode='same')
# Plot the loss results
data_type = 'loss'
fig, ax1 = plt.subplots(figsize=(10, 6))
x_list = list(range(1, ffo_result_mean[ydim].shape[0] + 1))
# Create a secondary y-axis
sns.lineplot(x=x_list, y=ffo_result_mean[ydim][:,1], label='FFO', ax=ax1, linewidth=2.5, zorder=10)
plt.fill_between(x_list, ffo_result_mean[ydim][:,1] - ffo_result_std[ydim][:,1], ffo_result_mean[ydim][:,1] + ffo_result_std[ydim][:,1], alpha=0.3, zorder=4)
sns.lineplot(x=x_list, y=cvxpylayer_result_mean[ydim][:,1], label='cvxpylayer', ax=ax1, linewidth=2.5, zorder=5)
plt.fill_between(x_list, cvxpylayer_result_mean[ydim][:,1] - cvxpylayer_result_std[ydim][:,1], cvxpylayer_result_mean[ydim][:,1] + cvxpylayer_result_std[ydim][:,1], alpha=0.3, zorder=4)
ax1.set_ylabel('Optimality gap', fontsize=28)
ax1.legend(loc='upper right', fontsize=28, frameon=False)
ax2 = ax1.twinx()
sns.barplot(x=x_list, y=grad_differences[ydim], label='grad_diff', ax=ax2, zorder=3)
ax2.set_ylabel('Gradient error', fontsize=28)
ax1.set_xlabel('Iteration', fontsize=28)
x_ticks = list(range(0, ffo_result_mean[ydim].shape[0], 50))
ax1.set_xticks(x_ticks)
ax1.set_xticklabels(x_ticks)
y1_min, y1_max = ax1.get_ylim()
y2_min, y2_max = ax2.get_ylim()
ax1.set_ylim(bottom=0)
ax2.set_ylim(bottom=0, top=1)
ax1.set_zorder(ax2.get_zorder() + 1)
ax1.patch.set_visible(False)
# plt.title('Convergence plot with y dimension = {}'.format(str(ydim)), fontsize=28)
plt.xlabel('Iteration', fontsize=28)
# handles, labels = plt.gca().get_legend_handles_labels()
# plt.legend(handles=handles[1:], labels=labels[1:], loc='upper right')
# Save figure
plt.tight_layout()
plt.savefig('figures/' + directory_name + '_{}.{}'.format(data_type, file_type))
plt.close()
# ================================== EXP 2 ===================================
# Computation cost analysis
ffo_result, cvxpylayer_result = {}, {}
ffo_result_mean, cvxpylayer_result_mean = {}, {}
ffo_result_std, cvxpylayer_result_std = {}, {}
eps = 0.01
ydim_list = list(range(100,1100,100))
directory_path = 'exp2/'
# directory_path = 'exp2_bilinear/'
seed_list = list(set(range(1,11,1))) #- set([1, 10,11,12,13,14,18,19,20]))
for ydim in ydim_list:
directory_name = directory_path + 'ydim{}'.format(ydim)
# Initialize the dictionary
ffo_result[ydim], cvxpylayer_result[ydim] = [], []
# Read the results
cvxpylayer_filename_list = ['cvxpylayer_eps{}_seed{}.txt'.format(eps, seed) for seed in seed_list]
ffo_filename_list = ['ffo_eps{}_seed{}.txt'.format(eps, seed) for seed in seed_list]
for filename in cvxpylayer_filename_list:
try:
df = pd.read_csv('results/' + directory_name + '/' + filename)
cvxpylayer_result[ydim].append(df.values[:,:3])
except:
print(ydim, filename)
for filename in ffo_filename_list:
try:
df = pd.read_csv('results/' + directory_name + '/' + filename, header=None, names=['iteration', 'loss', 'time', 'time1', 'time2'], skiprows=1)
ffo_result[ydim].append(df.values[:,:3])
except:
print(ydim, filename)
ffo_result_mean[ydim] = np.mean(ffo_result[ydim], axis=0)
ffo_result_std[ydim] = np.std(ffo_result[ydim], axis=0)
cvxpylayer_result_mean[ydim] = np.mean(cvxpylayer_result[ydim], axis=0)
cvxpylayer_result_std[ydim] = np.std(cvxpylayer_result[ydim], axis=0)
# Plot the time results
time_results = {'ffo': [], 'cvxpylayer': [], 'ydim': []}
for ydim in ffo_result.keys():
time_results['ffo'].append(np.mean(ffo_result_mean[ydim][:,2]))
time_results['cvxpylayer'].append(np.mean(cvxpylayer_result_mean[ydim][:,2]))
time_results['ydim'].append(ydim)
time_results = pd.DataFrame(time_results)
plt.figure(figsize=(10, 6))
g = sns.barplot(x='ydim', y='value', hue='variable', data=pd.melt(time_results, ['ydim']))
# plt.errorbar(time_results['ydim'], time_results['ffo'], yerr=ffo_result_std[ydim][:,2], fmt='none', color='black', capsize=5)
# plt.errorbar(time_results['ydim'], time_results['cvxpylayer'], yerr=cvxpylayer_result_std[ydim][:,2], fmt='none', color='black', capsize=5)
g.set_xlabel('Inner level dimension', fontsize=28)
g.set_ylabel('Time (s)', fontsize=28)
# Set tick label size
plt.xticks(fontsize=10)
plt.yticks(fontsize=10)
# Adjust the legend
handles, labels = g.get_legend_handles_labels()
labels = ['FFO', 'cvxpylayer'] # Remove "variable" from legend labels
g.legend(handles=handles, labels=labels, title='', fontsize=28, frameon=False)
plt.tight_layout()
g.figure.savefig('figures/' + directory_path + 'time_results.{}'.format(file_type))
plt.close()
# ================================== EXP 3 ===================================
# Epsilon analysis
# Convergence analysis
ffo_result, cvxpylayer_result = {}, {}
ffo_result_mean, cvxpylayer_result_mean = {}, {}
ffo_result_std, cvxpylayer_result_std = {}, {}
grad_differences = {}
eps = 0.01
ydim_list = [100, 200, 500] # list(range(100,1000,100))
directory_path = 'exp3/'
# directory_path = 'exp3_bilinear/'
seed_list = list(range(1,6,1))
for ydim in ydim_list:
# Initialize the dictionary
ffo_result[ydim], cvxpylayer_result[ydim] = {}, {}
ffo_result_mean[ydim], cvxpylayer_result_mean[ydim] = {}, {}
ffo_result_std[ydim], cvxpylayer_result_std[ydim] = {}, {}
grad_differences[ydim] = {}
# Plot the loss results
data_type = 'gradient_error'
fig, ax1 = plt.subplots(figsize=(10, 6))
# Some random seed didn't finish (for linear case)
if ydim == 100:
seed_list = [1,2,3,4,5]
elif ydim == 200:
seed_list = [1,3,4,5]
elif ydim == 500:
seed_list = [2,3,4,5]
else:
seed_list = [1,2,3,4,5]
for eps in [0.0001, 0.001, 0.01, 0.1, 1.0]:
ffo_result[ydim][eps], cvxpylayer_result[ydim][eps] = [], []
grad_differences[ydim][eps] = []
directory_name = directory_path + 'ydim{}'.format(ydim)
# Read the results
ffo_filename_list = ['ffo_eps{}_seed{}.txt'.format(eps, seed) for seed in seed_list]
ffo_grad_filename_list = ['ffo_eps{}_seed{}.pickle'.format(eps, seed) for seed in seed_list]
cvxpylayer_filename_list = ['cvxpylayer_eps{}_seed{}.txt'.format(eps, seed) for seed in seed_list]
cvxpylayer_gradient_filename_list = ['cvxpylayer_eps{}_seed{}.pickle'.format(eps, seed) for seed in seed_list]
for filename in ffo_filename_list:
df = pd.read_csv('results/' + directory_name + '/' + filename, header=None, names=['iteration', 'loss', 'time', 'time1', 'time2'], skiprows=1)
ffo_result[ydim][eps].append(df.values[:,:3] - df.values[:,:3].min(axis=0))
# for filename in cvxpylayer_filename_list:
# df = pd.read_csv('results/' + directory_name + '/' + filename)
# cvxpylayer_result[ydim][eps].append(df.values[:,:3])
ffo_result_mean[ydim][eps] = np.mean(ffo_result[ydim][eps], axis=0)
ffo_result_std[ydim][eps] = np.std(ffo_result[ydim][eps], axis=0)
cvxpylayer_result_mean[ydim][eps] = np.mean(cvxpylayer_result[ydim][eps], axis=0)
cvxpylayer_result_std[ydim][eps] = np.std(cvxpylayer_result[ydim][eps], axis=0)
x_list = list(range(1, ffo_result_mean[ydim][eps].shape[0] + 1))
# Create a secondary y-axis
sns.lineplot(x=x_list, y=ffo_result_mean[ydim][eps][:,1], label='FFO ' + r'$\alpha^2$' + '={}'.format(eps), ax=ax1, linewidth=2.5)
plt.fill_between(x_list, ffo_result_mean[ydim][eps][:,1] - ffo_result_std[ydim][eps][:,1], ffo_result_mean[ydim][eps][:,1] + ffo_result_std[ydim][eps][:,1], alpha=0.3)
ax1.set_ylabel('Optimality gap', fontsize=28)
ax1.legend(loc='upper right', fontsize=28, frameon=False)
ax1.set_xlabel('Iteration', fontsize=28)
x_ticks = list(range(0, ffo_result_mean[ydim][eps].shape[0], 50))
ax1.set_xticks(x_ticks)
ax1.set_xticklabels(x_ticks)
# plt.title('Convergence of different gradient accuracy', fontsize=28)
plt.xlabel('Iteration', fontsize=28)
# handles, labels = plt.gca().get_legend_handles_labels()
# plt.legend(handles=handles[1:], labels=labels[1:], loc='upper right')
# Save figure
plt.tight_layout()
plt.savefig('figures/' + directory_name + '_{}.{}'.format(data_type, file_type))
plt.close()