-
Notifications
You must be signed in to change notification settings - Fork 0
/
experiments.py
433 lines (352 loc) · 18.5 KB
/
experiments.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
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
import numpy as np
import matplotlib.pyplot as plt
import os
from utils.truthfulness import truthfulness_mechanism
from tqdm import tqdm
class Postprocessing:
def __init__(self):
self.colors = ['r', 'b', 'g', 'orange', 'pink', 'cyan', 'yellow', 'purple']
def get_lambda(self, data_path, alpha, num_agents, offset=2):
# check file existence
_, dataset, begin_path, end_path = self.path_check(data_path)
if dataset.lower() == "cifar10":
num_data = 3125
total_data = num_data * num_agents
elif dataset.lower() == "mnist":
num_data = 3750
total_data = num_data * num_agents
elif dataset.lower() == "ham":
num_data = 801
total_data = 10015
else:
ValueError("Incorrect Dataset Provided")
file = begin_path + '-run1' + end_path
agent_mcs = self.unpack_data(file, 1, num_agents, datatype='costs.log').flatten()
mc = agent_mcs[0]
lambda_coefficient = (1 / ((2 - alpha) * offset)) * num_data * total_data / (total_data - num_data)
lamb = lambda_coefficient * np.square(mc - offset / (2 * np.square(total_data)))
return lamb, mc, num_data, total_data, dataset
def penalty(self, data_path, alpha=2, offset=2, num_agents=16, h=201, runs=3, save_file=None):
# get lambda
alpha -= 1e-6
lamb, mc, num_data, total_data, dataset = self.get_lambda(data_path, alpha, num_agents, offset=offset)
# initialize eps
vary_data = np.linspace(num_data - 500, num_data + 500, h, endpoint=True)
true_m = int(np.sqrt(offset / (2 * mc)))
# compute penalty
penalties = lamb * np.square(mc / (2 * lamb) - offset / (4 * lamb * np.square(total_data)) + true_m - vary_data)
penalties_plus_cost = penalties + vary_data * mc
# plot truthfulness
plt.figure(figsize=(8, 6))
plt.plot(vary_data, penalties_plus_cost, color='tab:red')
plt.plot(true_m, penalties[np.argwhere(vary_data == true_m)[0]], 'h', color='tab:blue', markersize=8,
label='True Optimal Contribution')
plt.xlabel('Data Contributed $m_i$', fontsize=20, weight='bold')
plt.ylabel('Free-Riding Penalty + Data Costs', fontsize=20, weight='bold')
plt.legend(loc='best', fontsize=15)
plt.xticks(vary_data[::50])
plt.xlim([num_data - 500, num_data + 500])
plt.tick_params(axis='both', which='major', labelsize=15)
plt.grid(alpha=0.25)
# save figure
if save_file is None:
plt.show()
else:
sf = save_file + '-truthfulness-' + str(num_agents) + 'agents-' + dataset.lower() + '.jpg'
plt.savefig(sf, dpi=200)
def truthfulness_plots(self, data_paths, save_file=None, runs=3, h=121):
# loss or accuracy
dt = 'loss.log'
# initialize eps
epsilons = np.linspace(-0.3, 0.3, h, endpoint=True) * 100
ls = [':', '--', '-']
label_add = [' IID', ' N-IID (D-0.6)', ' N-IID (D-0.3)']
plt.figure(figsize=(8, 6))
for v, data_path in enumerate(data_paths):
# check file existence
_, dataset, begin_path, end_path = self.path_check(data_path)
if dataset.lower() == "cifar10":
num_data = 3125
elif dataset.lower() == "mnist":
num_data = 3750
elif dataset.lower() == "ham":
num_data = 801
else:
ValueError("Incorrect Dataset Provided")
y_mean_local, _, _, y_mean_fed, _, _, _, num_agents = self.get_loss_data(begin_path, end_path, runs, dt)
avg_local_loss = y_mean_local[-1]
avg_fed_loss = y_mean_fed[-1]
net_loss = avg_local_loss - avg_fed_loss
# get penalty term for free-riding (this is minimal because no free-riding at optimal ~ only tiny value)
alpha = 2 - 1e-6
lamb, mc, _, total_data, _ = self.get_lambda(data_path, alpha, num_agents, offset=2)
penalty = lamb * np.square(mc / (2 * lamb) - 2 / (4 * lamb * np.square(total_data)))
# get average benefit from participating in FACT
agent_net = np.empty((runs, num_agents))
other_agent_net = np.empty((runs, num_agents))
for run in range(1, runs + 1):
file = begin_path + '-run' + str(run) + end_path
fact_benefit = self.unpack_data(file, 3, num_agents, datatype='benefits.log')
agent_net[run - 1, :] = fact_benefit[0, :]
other_agent_net[run - 1, :] = fact_benefit[1, :]
# get truthfulness data
avg_agent_net = np.mean(agent_net, axis=0)
avg_other_agent_net = np.mean(other_agent_net, axis=0)
fbr = np.empty((num_agents, h))
fl = np.empty(num_agents)
for i in tqdm(range(num_agents)):
fl[i], fbr[i, :] = truthfulness_mechanism(mc, num_data, avg_local_loss, avg_agent_net[i], avg_other_agent_net[i],
num_agents, h=h, normal=True, agents=2000, rounds=100000)
avg_fbr = np.mean(fbr, axis=0) + penalty
# plot truthfulness
plt.plot(epsilons, avg_fbr, self.colors[0], label=label_add[v], linestyle=ls[v])
plt.xlabel('Percent (%) Added/Subtracted from True Cost $c_i$', fontsize=20, weight='bold')
plt.ylabel('Expected Improvement in Loss', fontsize=20, weight='bold')
if dataset.lower() != 'ham':
plt.legend(loc='upper left', fontsize=15)
# plt.legend(loc='upper right', fontsize=15)
plt.xlim([-30, 30])
plt.tick_params(axis='both', which='major', labelsize=15)
plt.grid(alpha=0.25)
# save figure
if save_file is None:
plt.show()
else:
sf = save_file + '-truthfulness-' + str(num_agents) + 'agents-' + dataset.lower() + '.jpg'
plt.savefig(sf, dpi=200)
def run_loss_histogram(self, data_path, save_file=None, loss=True, runs=3, h=121):
# loss or accuracy
dt = 'loss.log' if loss else 'acc-top1.log'
# check file existence
_, dataset, begin_path, end_path = self.path_check(data_path)
if dataset.lower() == "cifar10":
num_data = 3125
elif dataset.lower() == "mnist":
num_data = 3750
elif dataset.lower() == "ham":
num_data = 801
else:
ValueError("Incorrect Dataset Provided")
y_mean_local, _, _, y_mean_fed, _, _, epochs, num_agents = self.get_loss_data(begin_path, end_path, runs, dt)
avg_local_loss = y_mean_local[-1]
avg_fed_loss = y_mean_fed[-1]
net_loss = avg_local_loss - avg_fed_loss
# get penalty term for free-riding (this is minimal because no free-riding at optimal ~ only tiny value)
alpha = 2 - 1e-6
lamb, mc, _, total_data, _ = self.get_lambda(data_path, alpha, num_agents, offset=2)
penalty = lamb * np.square(mc / (2 * lamb) - 2 / (4 * lamb * np.square(total_data)))
# get average benefit from participating in FACT
agent_net = np.empty((runs, num_agents))
other_agent_net = np.empty((runs, num_agents))
for run in range(1, runs + 1):
file = begin_path + '-run' + str(run) + end_path
fact_benefit = self.unpack_data(file, 3, num_agents, datatype='benefits.log')
agent_net[run - 1, :] = fact_benefit[0, :]
other_agent_net[run - 1, :] = fact_benefit[1, :]
# get truthfulness data
avg_agent_net = np.mean(agent_net, axis=0)
avg_other_agent_net = np.mean(other_agent_net, axis=0)
fbr = np.empty((num_agents, h))
fl = np.empty(num_agents)
for i in tqdm(range(num_agents)):
fl[i], fbr[i, :] = truthfulness_mechanism(mc, num_data, avg_local_loss, avg_agent_net[i], avg_other_agent_net[i],
num_agents, h=h, normal=True, agents=2000, rounds=100000)
fact_loss = np.mean(fl) + penalty
# add for loop here over all the avg_fact_losses
# plot results
plt.figure(figsize=(8, 6))
plt.bar(["Local Training", 'FACT Training', 'Traditional FL'],
[avg_local_loss, fact_loss, avg_fed_loss],
color=['tab:red', 'tab:blue', 'tab:green'])
plt.ylabel('Expected Loss', fontsize=25, weight='bold')
plt.xticks(fontsize=17, weight='bold')
plt.yticks(fontsize=17, weight='bold')
if dataset.lower() == 'cifar10':
plt.ylim([0, 7])
elif dataset.lower() == 'mnist':
plt.ylim([0.001, 2])
plt.yscale("log")
plt.grid(alpha=0.25, axis='y')
plt.tick_params(axis='both', which='major', labelsize=16)
# save figure
if save_file is None:
plt.show()
else:
sf = save_file + '-loss-histogram-' + str(num_agents) + 'agents-' + dataset.lower() + '.jpg'
plt.savefig(sf, dpi=200)
def run_loss_plot(self, data_path, save_file=None, loss=True, runs=3):
# loss or accuracy
dt = 'loss.log' if loss else 'acc-top1.log'
# check file existence
_, dataset, begin_path, end_path = self.path_check(data_path)
y_mean_local, y_min_local, y_max_local, y_mean_fed, y_min_fed, y_max_fed, epochs, num_agents = (
self.get_loss_data(begin_path, end_path, runs, dt))
# plot results
plt.figure(figsize=(8, 6))
# local
plt.plot(range(1, epochs+1), y_mean_local, color='r', label='Local Training')
plt.fill_between(range(1, epochs+1), y_min_local, y_max_local, alpha=0.2, color='r')
# fed
plt.plot(range(1, epochs+1), y_mean_fed, color='b', label='Federated Training')
plt.fill_between(range(1, epochs+1), y_min_fed, y_max_fed, alpha=0.2, color='b')
plt.xlabel('Epochs', fontsize=20, weight='bold')
if loss:
plt.ylabel('Test Loss', fontsize=20, weight='bold')
else:
plt.ylabel('Test Accuracy', fontsize=20, weight='bold')
plt.xticks(fontsize=17, weight='bold')
plt.yticks(fontsize=17, weight='bold')
plt.legend(loc='best', fontsize=15)
if loss and dataset.lower() == 'mnist':
plt.ylim([0.01, 2.5])
plt.yscale("log")
elif loss and dataset.lower() == 'cifar10':
plt.ylim([0, 1.25 * 10**2])
plt.yscale("symlog")
elif not loss and dataset.lower() == 'cifar10':
plt.ylim([0.1, 0.8])
elif not loss and dataset.lower() == 'mnist':
plt.ylim([0.75, 1.])
plt.xlim([1, epochs])
plt.grid(alpha=0.25)
# save figure
if save_file is None:
plt.show()
else:
save_file = save_file + '-' + str(num_agents) + 'agents-' + dataset.lower()
save_file = save_file + '-loss.jpg' if loss else save_file + '-acc.jpg'
plt.savefig(save_file, dpi=200)
def get_epoch_data(self, data_path, datatype='fed-epoch-loss.log'):
# determine number of agents
string, location = [(i, i.find("devices")) for i in data_path.split("-") if i.find("devices") > 0][0]
num_agents = int(string[:location])
# determine the number of epochs
with open(data_path + '/r0-fed-epoch-loss.log') as f:
epochs = sum(1 for _ in f)
return self.unpack_data(data_path, epochs, num_agents, datatype), epochs, num_agents
def get_benefit_data(self, data_path):
# determine number of agents
string, location = [(i, i.find("devices")) for i in data_path.split("-") if i.find("devices") > 0][0]
num_agents = int(string[:location])
# load in benefit data
d = np.load(data_path + '/r0-expected-epsilon-benefit.npy')
benefit_data = np.empty((num_agents, len(d)))
benefit_data[0, :] = d
for r in range(1, num_agents):
benefit_data[r, :] = np.load(data_path + '/r' + str(r) + '-expected-epsilon-benefit.npy')
return self.unpack_data(data_path, 3, num_agents, datatype='benefits.log'), benefit_data, num_agents
def get_loss_data(self, begin_path, end_path, runs, data_type):
# extract benefit data
losses_fed = []
losses_local = []
for run in range(1, runs + 1):
file = begin_path + '-run' + str(run) + end_path
loss_data_local, _, _ = self.get_epoch_data(file, datatype='local-epoch-' + data_type)
loss_data_fed, epochs, num_agents = self.get_epoch_data(file, datatype='fed-epoch-' + data_type)
losses_fed.append(loss_data_fed[:, 0])
losses_local.append(np.mean(loss_data_local, axis=1))
losses_fed = np.stack(losses_fed, axis=0)
losses_local = np.stack(losses_local, axis=0)
# compute error bars over all three runs
y_mean_local, y_min_local, y_max_local = self.generate_confidence_interval(losses_local)
y_mean_fed, y_min_fed, y_max_fed = self.generate_confidence_interval(losses_fed)
return y_mean_local, y_min_local, y_max_local, y_mean_fed, y_min_fed, y_max_fed, epochs, num_agents
def generate_confidence_interval(self, ys, number_per_g=30, number_of_g=1000, low_percentile=1, high_percentile=99):
means = []
mins = []
maxs = []
for i, y in enumerate(ys.T):
y = self.bootstrapping(y, number_per_g, number_of_g)
means.append(np.mean(y))
mins.append(np.percentile(y, low_percentile))
maxs.append(np.percentile(y, high_percentile))
return np.array(means), np.array(mins), np.array(maxs)
def plot_ci(self, x, y, num_runs, num_dots, mylegend, ls='-', lw=3, transparency=0.2):
assert (x.ndim == 1)
assert (x.size == num_dots)
assert (y.ndim == 2)
assert (y.shape == (num_runs, num_dots))
y_mean, y_min, y_max = self.generate_confidence_interval(y)
plt.plot(x, y_mean, 'o-', label=mylegend, linestyle=ls, linewidth=lw) # , label=r'$\alpha$={}'.format(alpha))
plt.fill_between(x, y_min, y_max, alpha=transparency)
return
@staticmethod
def path_check(data_path):
# ensure data files exist
test_file = data_path + '/r0-fed-epoch-loss.log'
if not os.path.isfile(test_file):
print(test_file)
raise Exception(f"Incorrect Path Provided")
# determine which dataset and which truthfulness method
method = 'Random Mechanism'
if data_path.lower().find('mnist') > -1:
dataset = 'MNIST'
elif data_path.lower().find('cifar') > -1:
dataset = 'Cifar10'
else:
dataset = 'HAM'
# extract all runs
split_paths = data_path.split("-run")
begin_path = split_paths[0]
end_path = split_paths[1][1:]
return method, dataset, begin_path, end_path
@staticmethod
def unpack_data(directory_path, epochs, num_workers, datatype):
directory = os.path.join(directory_path)
if not os.path.isdir(directory):
raise Exception(f"custom no directory {directory}")
data = np.zeros((epochs, num_workers))
for root, dirs, files in os.walk(directory):
for file in files:
if file.endswith(datatype):
j = int(file.split('-')[0][1:])
with open(directory_path + '/' + file, 'r') as f:
i = 0
for line in f:
itms = line.strip().split('\n')[0]
data[i, j] = float(itms)
i += 1
return data
@staticmethod
def bootstrapping(data, num_per_group, num_of_group):
new_data = np.array([np.mean(np.random.choice(data, num_per_group, replace=True)) for _ in range(num_of_group)])
return new_data
if __name__ == '__main__':
# iid
cifar10_random_path_iid = 'output/CIFAR10/fact-random-sandwich-uniform-cost-run1-cifar10-16devices'
mnist_random_path_iid = 'output/MNIST/fact-random-sandwich-uniform-cost-run1-mnist-16devices'
ham_random_path_iid = 'output/HAM/fact-sandwich-uniform-cost-run1-ham10000-10devices'
# noniid D-0.3
cifar10_random_path_noniid3 = 'output/CIFAR10/fact-random-sandwich-uniform-cost-noniid-0.3-run1-cifar10-16devices'
mnist_random_path_noniid3 = 'output/MNIST/fact-random-sandwich-uniform-cost-noniid-0.3-run1-mnist-16devices'
# noniid D-0.6
cifar10_random_path_noniid6 = 'output/CIFAR10/fact-random-sandwich-uniform-cost-noniid-0.6-run1-cifar10-16devices'
mnist_random_path_noniid6 = 'output/MNIST/fact-random-sandwich-uniform-cost-noniid-0.6-run1-mnist-16devices'
# combined
combined_cifar = [cifar10_random_path_iid, cifar10_random_path_noniid6, cifar10_random_path_noniid3]
combined_mnist = [mnist_random_path_iid, mnist_random_path_noniid6, mnist_random_path_noniid3]
# initialize postprocessing
pp = Postprocessing()
# loss plots
## pp.run_loss_plot(ham_random_path_iid, save_file='iid', runs=3, loss=False)
## pp.run_loss_plot(ham_random_path_iid, save_file='iid', runs=3, loss=True)
# pp.run_loss_plot(mnist_random_path_iid, save_file='iid', runs=3, loss=False)
# pp.run_loss_plot(cifar10_random_path_iid, save_file='iid', runs=3, loss=True)
# pp.run_loss_plot(mnist_random_path_noniid6, save_file='noniid6', runs=3, loss=False)
# pp.run_loss_plot(cifar10_random_path_noniid6, save_file='noniid6', runs=3, loss=True)
# pp.run_loss_plot(mnist_random_path_noniid3, save_file='noniid3', runs=3, loss=False)
# pp.run_loss_plot(cifar10_random_path_noniid3, save_file='noniid3', runs=3, loss=True)
# loss histogram
# pp.run_loss_histogram(ham_random_path_iid, save_file='iid', runs=3)
pp.run_loss_histogram(cifar10_random_path_iid, save_file='iid', runs=3)
pp.run_loss_histogram(cifar10_random_path_noniid6, save_file='noniid6')
pp.run_loss_histogram(cifar10_random_path_noniid3, save_file='noniid3')
pp.run_loss_histogram(mnist_random_path_iid, save_file='iid', runs=3)
pp.run_loss_histogram(mnist_random_path_noniid6, save_file='noniid6')
pp.run_loss_histogram(mnist_random_path_noniid3, save_file='noniid3')
# penalty for using suboptimal data contributions
## pp.penalty(ham_random_path_iid, save_file='penalty', offset=1)
# truthfulness plots
## pp.truthfulness_plots([ham_random_path_iid], save_file='vary-dist', runs=3)
pp.truthfulness_plots(combined_cifar, save_file='vary-dist', runs=3)
pp.truthfulness_plots(combined_mnist, save_file='vary-dist', runs=3)