-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathplot.py
185 lines (145 loc) · 5.54 KB
/
plot.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
import json
import os
import numpy as np
import matplotlib.pyplot as plt
import sys
from collections import defaultdict
from matplotlib.ticker import ScalarFormatter
COLOR_MAP = {
'FedAvg (S=100)': 'red',
'FedAvg (S=10)': '#FF6EB4',
'FedAvg (S=25)': '#6E6E6E',
'FedAvg (S=50)': 'forestgreen',
'FedAvg-IS': 'cornflowerblue',
'biased FedAvg': 'darkorange',
'MIFA': '#CD00CD',
}
def opt_reader(options):
"""
Give Option a better algorithm name. used for plotting
"""
if options['algo']=='fedavg':
return 'FedAvg (S='+str(options['clients_per_round'])+')'
if options['algo']=='sgd':
if not options['importance_sampling']:
return 'biased FedAvg'
else:
return 'FedAvg-IS'
if options['algo']=='fdu':
return 'MIFA'
if options['algo']=='fdu_no_wait':
return 'MIFA (0)'
def data_preprocess(options, logs):
if 'fedavg' in options['algo']:
time_steps = sorted([int(x) for x in logs.keys()])
for i in range(min(time_steps) + 1, options['num_round']):
if i not in time_steps:
logs[str(i)] = logs[str(i-1)]
def beautiful_print_report(report_dict):
for algo, res in report_dict.items():
gap = res['acc'] - report_dict['MIFA']['acc']
print("{} acc:{:.2f} std:{:.2f} gap:{:.2f} \n".format(algo, res['acc'], res['std'], gap))
def smooth_curve(y_arr, n=2):
length = y_arr.shape[0]
ret = np.zeros(length)
ret[:n] = y_arr[:n]
ret[-n:] = y_arr[-n:]
for idx in range(n, length - n):
ret[idx] = np.mean(y_arr[(idx - n) : (idx + n + 1)])
return(ret)
ROOT_DIR = sys.argv[1]
"""
assume that the result dir is organized as
results/experiment_id/log.json
results/experiment_id/options.json
"""
experiment_list = []
for roots, dirs, files in os.walk(ROOT_DIR):
if 'options.json' in files:
experiment_list.append(roots)
print("exp list length", len(experiment_list))
log = defaultdict(list)
for exp_id in experiment_list:
log_file = os.path.join(exp_id, 'log.json')
opt_file = os.path.join(exp_id, 'options.json')
with open(log_file) as f:
log_dict = json.load(f)
with open(opt_file) as f:
opt_dict = json.load(f)
del opt_dict['seed']
del opt_dict['result_dir']
del opt_dict['device']
log[repr(opt_dict)].append(log_dict) # group experiments of the same setting but different random seeds
data_preprocess(opt_dict, log_dict)
avail_plots = ['train_loss','train_acc','test_loss','test_acc']
target = avail_plots[int(sys.argv[2])]
print('plotting ', target)
fig, ax = plt.subplots(figsize = (10, 8))
handler = {}
myfontdict2 = {'family':'Times New Roman','size':20}
report_dict = defaultdict(dict)
for options, logs in log.items():
options = eval(options)
total_round = options['num_round']
total_round = 2000
first_times = [min( [int(x) for x in log_dict.keys()] ) for log_dict in logs]
max_first_time = max(first_times)
time_steps = range(max_first_time + 1, total_round + 1)
alg_name = opt_reader(options)
ys = [[log_dict[str(t-1)][target] for t in time_steps] for log_dict in logs]
ys = np.array(ys)
if target=='test_acc':
ys = ys*100
y_mean = np.mean(ys,0)
y_std = np.std(ys,0)
alg_name = opt_reader(options)
color= COLOR_MAP[alg_name]
y_mean = smooth_curve(y_mean, n=2)
y_std = smooth_curve(y_std, n=2)
handler[alg_name] = plt.plot(time_steps, y_mean, color)[0]
report_dict[alg_name]['acc'] = y_mean[-1]
report_dict[alg_name]['std'] = y_std[-1]
plt.fill_between(time_steps, y_mean-y_std, y_mean+y_std, alpha=0.3, color =color)
report_dict = dict(sorted(report_dict.items(), key=lambda item: item[1]['acc'], reverse=True))
beautiful_print_report(report_dict)
assert sys.argv[3] in ['cifar', 'mnist']
dataset = sys.argv[3]
if target=='train_loss':
plt.xlabel("communication round", myfontdict2)
plt.ylabel("training loss", myfontdict2)
plt.yscale('log')
# scaling for MNIST
if dataset == 'mnist':
plt.ylim(0.4,1.8)
ax.set_yticks([0.4,0.6,0.8,1,1.2,1.4])
ax.yaxis.set_major_formatter(ScalarFormatter())
#ax.yaxis.set_minor_formatter(ScalarFormatter())
else:
plt.ylim(1.8,2.59)
ax.set_yticks([2.0,2.1,2.2,2.3,2.4])
ax.yaxis.set_major_formatter(ScalarFormatter())
ax.yaxis.set_minor_formatter(ScalarFormatter())
label = sorted(handler.keys())
handler_content = [handler[x] for x in label]
myfontdict2 = {'family':'Times New Roman','size':20}
ax.tick_params(axis='both', which='major', labelsize=20)
ax.tick_params(axis='both', which='minor', labelsize=20)
#plt.legend(handler_content, label, prop=myfontdict2)
plt.savefig("./plot/new_{}_{}_{}_train_loss.pdf".format(dataset,part,net), bbox_inches='tight')
if target=='test_acc':
plt.xlabel("communication round", myfontdict2)
plt.ylabel("test acc (%)", myfontdict2)
plt.yscale('linear')
if dataset == 'mnist':
plt.ylim(50,91)
else:
plt.ylim(15, 38)
plt.yticks([15, 20, 25, 30, 35])
ax.yaxis.set_major_formatter(ScalarFormatter())
ax.yaxis.set_minor_formatter(ScalarFormatter())
label = sorted(handler.keys())
handler_content = [handler[x] for x in label]
ax.tick_params(axis='both', which='major', labelsize=20)
ax.tick_params(axis='both', which='minor', labelsize=20)
#plt.legend(handler_content, label, prop=myfontdict2)
plt.savefig("./plot/{}_{}.pdf".format(dataset, avail_plots[int(sys.argv[2])]), bbox_inches='tight')