/
slave.py
216 lines (190 loc) · 8.37 KB
/
slave.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
#############################################################################
#! /usr/bin/env python
"""
Slave for distributed optimization
"""
import socket
import select
import sys
import time
import argparse
import numpy as np
from scipy.sparse import csr_matrix
from scipy.sparse.linalg import norm
from sklearn.datasets import load_svmlight_file
from communication import send, receive
from functions import prox_r, f_grad
from data_processer import get_data
PRINT_SUFFIX = ''
def slave_print(*args, **kw_args):
print('Slave' + PRINT_SUFFIX + ':', end= ' ')
print(*args, **kw_args)
class Slave(object):
def __init__(self, ip=None, port=8888, path=''):
# Quit flag
self.path = path
if ip is None:
with open(path + 'ip', 'r') as f:
ip = f.read()
self.flag = False
self.port = int(port)
self.x = csr_matrix(0)
self.x_ave = csr_matrix(0)
self.alpha = csr_matrix(0)
self.alpha_ave = csr_matrix(0)
self.n = 0
self.n_i = 0
self.algo = None
self.i = None
# Connect to server at port
try:
self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.sock.connect((ip, self.port))
slave_print ('Connected to server@%d' % self.port)
# Send slave's name...
data = receive(self.sock)
# Contains slave address, set it
addr = data[0].split('SLAVE: ')[1]
self.i = data[1]
global PRINT_SUFFIX
PRINT_SUFFIX = ' ' + str(self.i)
except socket.error as e:
slave_print ('Could not connect to server @%d' % self.port)
sys.exit(1)
def wait_until_get_data(self):
while True:
try:
inputready, outputready, exceptrdy = select.select([0, self.sock], [],[])
except select.error as e:
break
except socket.error as e:
break
for i in inputready:
data = receive(self.sock)
if data is None:
slave_print('Something went wrong')
self.flag = True
break
can_read, n_features = data
if can_read == True:
slave_print('Reading my part of the data with', n_features, 'features...')
self.A, self.b = load_svmlight_file(self.path + 'slave' + str(self.i) + '_data', n_features=n_features, zero_based=True)
self.b = csr_matrix(self.b).T
self.n_i = self.b.shape[0]
return
def wait_until_get_parameters(self):
parameters_not_known = True
while parameters_not_known:
try:
inputready, outputready, exceptrdy = select.select([0, self.sock], [],[])
except select.error as e:
break
except socket.error as e:
break
for i in inputready:
data = receive(self.sock)
if data is None:
slave_print('Something went wrong')
self.flag = 1
parameters_not_known = False
break
slave_print('Receiving parameters...')
self.x_ave, self.M, self.gamma, self.l2, self.l1, self.n, self.algo, data_name = data
slave_print('Received parameters: M = %d, l2 = %3f, l1 = %3f, n = %d, algo = %s, dataset is %s'
% (self.M, self.l2, self.l1, self.n, self.algo, data_name))
self.x = self.x_ave
parameters_not_known = False
return
def serve(self):
self.wait_until_get_data()
self.wait_until_get_parameters()
self.alpha = csr_matrix(np.zeros(self.x.shape[0])).T
self.alpha_ave = csr_matrix(np.zeros(self.x.shape[0])).T
slave_print('Start optimizing...')
it = 0
while not self.flag:
it += 1
if it == 1:
start = time.time()
if self.algo in ['asynch_ave', 'daga']:
grad = f_grad(self.x, self.A, self.b, self.l2)
delta_grad = (grad - self.alpha) * (self.n_i / self.n)
if self.algo == 'asynch_ave':
delta_x = -self.gamma * (self.alpha_ave + delta_grad)
else:
delta_x = -self.gamma * (self.alpha_ave + delta_grad * (self.n / self.n_i))
elif self.algo == 'daga2':
z = prox_r(self.x_ave, self.gamma, self.l1)
grad = f_grad(z, self.A, self.b, self.l2)
delta_grad = (grad - self.alpha) * (self.n_i / self.n)
x_new = z - self.gamma * (self.alpha_ave + delta_grad * (self.n / self.n_i))
delta_x = (x_new - self.x_ave)
self.x = x_new
elif self.algo == 'daga3':
x_new = z - self.gamma * grad
delta_x = (x_new - self.x) * self.n_i / self.n - self.gamma * self.alpha_ave
self.x = x_new
if self.algo in ['asynch_ave', 'daga', 'daga2']:
self.alpha = grad
data = [delta_x, delta_grad]
elif self.algo in ['synch_gd', 'asynch_gd']:
delta = csr_matrix(np.zeros(self.A.shape[1])).T
p = 1
for i in range(p):
z = prox_r(self.x_ave + delta, self.gamma, self.l1)
grad = f_grad(z, self.A, self.b, self.l2)
x_new = z - self.gamma * grad
delta += (self.n_i / self.n) * (x_new - self.x)
self.x = x_new
data = delta
if it == 1:
dif = time.time() - start
n_passes = p if self.algo == 'asynch_gd' else 1
slave_print('It takes', dif, 'seconds to make', n_passes, 'pass(es) through the data')
start = time.time()
send(self.sock, data)
if it == 1:
end = time.time()
slave_print('It takes', end - start, 'seconds to send an update')
not_updated = True
while not_updated:
try:
inputready, outputready, exceptrdy = select.select([0, self.sock], [],[])
for inp in inputready:
if inp == self.sock:
data = receive(self.sock)
not_updated = False
if data is None:
slave_print('Shutting down...')
self.flag = True
break
elif type(data) is str and data == 'Change algorithm':
self.wait_until_get_parameters()
slave_print('Start optimizing...')
continue
elif type(data) is str and data == 'Terminate':
slave_print('Terminating.')
self.flag = True
break
if self.algo in ['asynch_ave', 'daga', 'daga2']:
self.x, self.alpha_ave = data
break
else:
self.x_ave = data
break
except KeyboardInterrupt:
slave_print('Interrupted.')
self.sock.close()
break
# crash = np.random.binomial(1, 0.01)
# if crash:
# time.sleep(2)
self.sock.close()
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Serve as a worker for distritued optimization')
parser.add_argument('--ip', action='store', help='ip address of the server', default=None)
parser.add_argument('--port', action='store', dest='port', default = 8888, type=int, help='Server\'s port')
parser.add_argument('--path', action='store', default='', help='Path to the data')
results = parser.parse_args()
slave = Slave(results.ip, results.port, results.path)
slave.serve()