/
hyperparameter.py
executable file
·169 lines (150 loc) · 5.47 KB
/
hyperparameter.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
#!/usr/bin/python2
import math
import time
import numpy as np
import Data
import sys
sys.path.append('EM/')
sys.path.append('EP/')
import EM_net
import EP_net
def compute_result(y,result,ccc):
if (ccc):
e1 = np.mean(y)
e2 = np.mean(result)
a = y - e1
b = result - e2
res = 2*np.mean(a*b)/(np.var(y) + np.var(result) + (e1 - e2)**2)
else:
res = np.sqrt(np.mean((y - result)**2))
return res
def ep_run(X_train,y_train,X_test,y_test,n_hidden_units,lam=0.1,var_prior=1,ccc=False):
net = EP_net.EP_net(X_train, y_train,
[n_hidden_units ],lam,var_prior)
net.train(X_train,y_train,10)
m, v, v_noise = net.predict(X_test)
test_res = compute_result(y_test,m,ccc)
return test_res
def em_run(X_train,y_train,X_test,y_test,n_hidden_units,lam=0.1,eta=1,a=0.01,ccc=False):
em = EM_net.EM_net(X_train,y_train, n_hidden_units,lam,eta,a)
em.sgd(X_train,y_train,n_epochs=20)
output = em.sgd_predict(X_test)
test_res = compute_result(y_test,output,ccc)
return test_res
def hyper_search(X,y,X_dev=None,y_dev=None,ccc=False):
if X_dev==None:
dataset = Data.partition(X,y)
X_train = dataset['X_train']
y_train = dataset['y_train']
X_dev = dataset['X_dev']
y_dev = dataset['y_dev']
else:
X_train = X
y_train = y
# Find Optimal Hyperparameter Setting
lam_arr = [0.01,0.05,0.1,1,10]
a_arr = [0,0.001,0.01,0.1,0.5,1]
eta_arr = [0.01,0.1,0.3,0.5,0.7,0.9,1,1.1]
n_arr = [10,20,30,50]
var_prior_arr = [0.1,0.5,0.7,1,1.1,1.5,2]
# lam_arr = [0.02,0.03,0.04]
# a_arr = [0,0.01]
# eta_arr = [0.0001]
# n_arr = [50,100,150]
# var_prior_arr = [0.1,3]
best_ep = 1e6
best_em = 1e6
if ccc:
best_ep = -1e6
best_em = -1e6
params = {}
for lam in lam_arr:
for n in n_arr:
###############################FOR EP################################################
for var_prior in var_prior_arr:
err = ep_run(X_train,y_train,X_dev,y_dev,n,lam=lam,var_prior=var_prior,ccc=ccc)
if(ccc and err > best_ep )or \
(not ccc and err < best_ep):
best_ep = err
params['lam'] = lam
params['n'] = n
params['var_prior'] = var_prior
###############################FOR EM################################################
for eta in eta_arr:
for a in a_arr:
err = em_run(X_train,y_train,X_dev,y_dev,n,
lam=lam,eta=eta,a=a,ccc=ccc)
if(ccc and err > best_em )or \
(not ccc and err < best_em):
best_em = err
params['lam_em'] = lam
params['n_em'] = n
params['eta'] = eta
params['a'] = a
# print params
print "best EP error: " + str(best_ep)
print "best EM error: " + str(best_em)
print "best params"
print params
# #################### We load artificial data from an RBFNN ########################
# # n_dim = 10
# # n_nodes = 10
# # n_pts = 100
# # c = np.random.rand(n_nodes,n_dim)*2
# # w = np.random.rand(n_nodes,1)*3
# # # generate random inputs with gaussian noise
# # X = (np.random.rand(n_pts,n_dim) - 0.5)*4 + np.random.randn(n_pts,n_dim)
# # y = []
# # for x_in in X:
# # #rbfs = np.exp(-0.1*np.sum((x_in - c)**2,axis=1))
# # #y.append(np.sum(w*rbfs))
# # sins = np.sin(2*x_in)
# # y.append(np.sum(2*sins))
# # y = np.array(y + 1*np.random.randn(n_pts))
# # eta = 1.1
# # a = 0.1
# # n_hidden_units = 50
# # print 'Artificial Optimal HyperParameter'
# ################### We load the boston housing dataset ###########################
# data = np.loadtxt('boston_housing.txt')
# X = data[ :, range(data.shape[ 1 ] - 1) ]
# y = data[ :, data.shape[ 1 ] - 1 ]
# print 'Boston Housing Optimal HyperParameter'
# ################### We load concrete dataset ######################################
# csv = np.genfromtxt ('concrete.csv', delimiter=",",skip_header=1)
# X = csv[ :, range(csv.shape[ 1 ] - 1) ]
# y = csv[ :, csv.shape[ 1 ] - 1 ]
# print 'Concrete Optimal HyperParameter'
# hyper_search(X,y)
# ##################### We load forestfires dataset #################################
# csv = np.genfromtxt ('forestfires.csv', delimiter=",",skip_header=1)
# ind = range(csv.shape[ 1 ] - 1)
# ind = [x for x in ind if (x != 2 and x != 3)]
# X = csv[ :,ind]
# y = csv[ :, csv.shape[ 1 ] - 1 ]
# for i in range(len(y)):
# if y[i] > 0:
# y[i] = np.log(y[i])
# print 'Forestfires Optimal HyperParameter'
# hyper_search(X,y)
# ################ Artificial 1D dataset ##########################################
# # num_train = 1000
# # def generate_xy(rng,num,noise=True):
# # x_pts = np.linspace(-rng,rng,num=num)
# # X = np.array([x_pts]).T
# # if (noise):
# # y = 3*np.cos(x_pts/9) + 2*np.sin(x_pts/15) + 0.1*np.random.randn(num)
# # else:
# # y = 3*np.cos(x_pts/9) + 2*np.sin(x_pts/15)
# # return(X,y)
# # rng = 80
# # X,y = generate_xy(rng,num_train)
# # hyper_search(X,y)
# ##################### We load the word music dataset ###############################
# csv = np.genfromtxt ('music.csv', delimiter=",",skip_header=1)
# X = csv[ :, range(csv.shape[ 1 ] - 2) ]
# y = csv[ :, csv.shape[ 1 ] - 1 ]
# print 'World Music Optimal HyperParameter'
# hyper_search(X,y)
# y = csv[ :, csv.shape[ 1 ] - 2 ]
# hyper_search(X,y)