/
NeuralNetBasicOSI.py
191 lines (157 loc) · 6.89 KB
/
NeuralNetBasicOSI.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
import numpy as np
from sklearn.base import BaseEstimator,ClassifierMixin
from sklearn.utils import check_random_state, array2d
from Utils import mean_squared_error as cost
from itertools import tee, izip, product
from scipy.special import expit as sigmoid
from scipy.stats import linregress
from sklearn import cross_validation as cv
from collections import deque
from operator import attrgetter
from sklearn.preprocessing import label_binarize, MultiLabelBinarizer
from NeuralNetBasicSwarm import BasicSwarm
class BasicOSI(BaseEstimator, ClassifierMixin):
def __init__(self, n_hidden, random_state=None, num_particles=20,
min_weight=-3, max_weight=3, window=10, validation_size=0.25,
c1=1.49445, c2=1.49445, w=0.729, min_v=-2, max_v=2, verbose=False):
self.n_hidden = n_hidden
self.random_state = random_state
self.num_particles = num_particles
self.min_weight = min_weight
self.max_weight = max_weight
self.min_v = min_v
self.max_v = max_v
self.c1 = c1
self.c2 = c2
self.w = w
self.window = window
self.validation_size = validation_size
self.verbose = verbose
self.mlb = MultiLabelBinarizer()
self.cost = cost
@staticmethod
def pairwise(iterable):
"s -> (s0,s1), (s1,s2), (s2, s3), ..."
a, b = tee(iterable)
next(b, None)
return izip(a, b)
def construct_paths(self):
layers = list()
layers.append(self.n_in)
layers.extend([x for x in self.n_hidden])
layers.append(self.n_out)
# construct network
tmp = [range(x) for x in layers]
tmp = list(product(*tmp))
paths = [list(self.pairwise(i)) for i in tmp]
# add bias
for i in xrange(len(layers) - 1):
tmp = [range(x) if j > 0 else [x] for j, x in enumerate(layers[i:])]
tmp = list(product(*tmp))
tmp = [[None] * i + list(self.pairwise(j)) for j in tmp]
paths.extend(tmp)
return paths
def reconstruct_gvn(self, W):
w = W[:]
# argmin(s_i)
for i in xrange(len(self.W_swarms)):
for s in self.W_swarms[i]:
best = min(s, key=attrgetter('s_best_score'))
if best.s_best_score is not np.inf:
p = s[0].path[i]
w[i][p] = best.s_best_weight[i]
return w
def init_params(self, X, y):
self.random_state = check_random_state(self.random_state)
self.classes_, y = np.unique(y, return_inverse=True)
X = array2d(X)
_, self.n_in = X.shape
self.n_out = np.unique(y).shape[0]
def init_network(self):
# construct the global network, +1 represents the bias layer
W = list()
for i in xrange(len(self.n_hidden)):
if i == 0:
W.append(self.random_state.uniform(self.min_weight, self.max_weight, (self.n_in + 1, self.n_hidden[i])))
else:
W.append(self.random_state.uniform(self.min_weight, self.max_weight, (self.n_hidden[i-1] + 1, self.n_hidden[i])))
W.append(self.random_state.uniform(self.min_weight, self.max_weight, (self.n_hidden[-1] + 1, self.n_out)))
# initialize the swarms
self.swarms = [BasicSwarm(self.n_in, self.n_hidden, self.n_out, path,
num_particles=self.num_particles, random_state=self.random_state,
min_weight=self.min_weight, max_weight=self.max_weight,
min_v=self.min_v, max_v=self.max_v)
for path in self.paths]
return W
def fit(self, X, y):
self.init_params(X, y)
self.paths = self.construct_paths()
num = len(self.paths[0])
swarm_paths = [sorted(list(set([s[i] for s in self.paths if s[i] is not None]))) for i in xrange(num)]
W = self.init_network()
self.W_swarms = [[[s for s in self.swarms if s.path[j] == i] for i in swarm_paths[j]] for j in xrange(num)]
X_train, X_valid, y_train, y_valid = cv.train_test_split(X, y, test_size=self.validation_size,
random_state=self.random_state)
# binarize true values
if len(self.classes_) > 2:
y_train = label_binarize(y_train, self.classes_)
y_valid = label_binarize(y_valid, self.classes_)
else:
y_train = self.mlb.fit_transform(label_binarize(y_train, self.classes_))
y_valid = self.mlb.fit_transform(label_binarize(y_valid, self.classes_))
j = 0
tmp = [1e3 - float(x * 1e3)/self.window for x in xrange(self.window)]
window = deque(tmp, maxlen=(self.window * 5))
self.num_evals = 0
best_score = np.inf
if self.verbose:
print "Fitting network {0}-{1}-{2} with {3} paths".format(self.n_in, self.n_hidden, self.n_out, len(self.swarms))
while True:
j += 1
for s in self.swarms:
for p_index in xrange(self.num_particles):
self.num_evals += 1
# evaluate each swarm
score = s.evaluate(W, X_train, y_train, p_index)
# reconstruct gvn
Wn = self.reconstruct_gvn(W)
# update
s.update(self.w, self.c1, self.c2, p_index)
# evaluate gvn
y_pred = self.forward(Wn, X_valid)
score = self.cost(y_valid, y_pred)
if score < best_score:
W = Wn[:]
best_score = score
window.append(best_score)
r = linregress(range(self.window), list(window)[-self.window:])
if self.verbose:
print j, best_score
if r[0] >= 0 or best_score < 1e-3:
self.W = W
self.num_generations = j
return self
def forward(self, W_in, X):
# construct network
tmp = X
for i in xrange(len(W_in)):
if i == 0:
W, W_b = W_in[i][:self.n_in], W_in[i][self.n_in:]
else:
W, W_b = W_in[i][:self.n_hidden[i-1]], W_in[i][self.n_hidden[i-1]:]
tmp = sigmoid(np.dot(tmp, W) + W_b)
return tmp
def decision_function(self, X):
return self.forward(self.W, X)
def predict(self, X):
scores = self.decision_function(X)
results = np.argmax(scores, axis=1)
return self.classes_[results]
def predict_proba(self, X):
return self.decision_function(X)
def predict_log_proba(self, X):
return np.log(self.predict_proba(X))
def get_num_evals(self):
return np.sum([s.get_num_evals() for s in self.swarms]) + self.num_evals
def get_num_generations(self):
return self.num_generations