/
main.py
139 lines (106 loc) · 4.6 KB
/
main.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
#!/usr/bin/env python
# encoding: utf-8
"""
Experiments for Homework #2 in in Yann LeCun's Machine Learning class a NYU
All original code by Daniel Foreman-Mackey
Most of the skeleton code has been loosely "ported" from the provided LUSH code.
"""
import sys
import numpy as np
from backprop import Machine, modules
from dataset import Dataset
data = None
# data = Dataset('dataset/spambase.data')
def run_machine(mach, save=True, **kwargs):
N = mach.nparams
print "Running machine w/ architecture"
print "\t", mach
print "and %d parameters"%(N)
print "Initial L = %.4f, f_error = %.4f on training set"%(mach.test(training_set=True))
print "Initial L = %.4f, f_error = %.4f on test set"%(mach.test())
mach.train(**kwargs)
print "Final L = %.4f, f_error = %.4f on training set"%(mach.test(training_set=True))
testing = mach.test()
print "Final L = %.4f, f_error = %.4f on test set\n"%(testing)
if save:
f = open('results.dat', 'a')
f.write("%d %.4f\n"%(N,testing[1]))
f.close()
def logistic_regression():
mach = Machine(data)
mach.add_module(modules.LinearModule, kwargs={'dim_out': data.nclass})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SigmoidModule)
mach.add_module(modules.EuclideanModule)
run_machine(mach, eta=0.0025, decay=0.0025, tol=5.25e-5)
def single_layer():
mach = Machine(data)
mach.add_module(modules.LinearModule, kwargs={'dim_out': data.nclass})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SoftMaxModule)
mach.add_module(modules.CrossEntropyModule)
run_machine(mach, eta=0.001, decay=0.0001, tol=1.25e-2)
def double_layer(nhidden=80, eta=0.001, decay=0.0001, tol=1.25e-2):
mach = Machine(data)
mach.add_module(modules.LinearModule, kwargs={'dim_out': nhidden})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SigmoidModule)
mach.add_module(modules.LinearModule, kwargs={'dim_out': data.nclass})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SoftMaxModule)
mach.add_module(modules.CrossEntropyModule)
run_machine(mach, eta=eta, decay=decay, tol=tol)
def triple_layer(nhidden=80, eta=0.001, decay=0.0001, tol=1.25e-2):
mach = Machine(data)
mach.add_module(modules.LinearModule, kwargs={'dim_out': nhidden})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SigmoidModule)
mach.add_module(modules.LinearModule, kwargs={'dim_out': nhidden})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SigmoidModule)
mach.add_module(modules.LinearModule, kwargs={'dim_out': data.nclass})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SoftMaxModule)
mach.add_module(modules.CrossEntropyModule)
run_machine(mach, eta=eta, decay=decay, tol=tol)
def rbf_hybrid(nhidden=80, eta=0.001, decay=0.0001, tol=1.25e-2):
templates = np.random.randn(nhidden*data.nclass).reshape(nhidden, data.nclass)
mach = Machine(data)
mach.add_module(modules.LinearModule, kwargs={'dim_out': nhidden})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SigmoidModule)
mach.add_module(modules.RBFModule, args=[templates])
mach.add_module(modules.BiasModule)
mach.add_module(modules.SoftMaxModule)
mach.add_module(modules.CrossEntropyModule)
run_machine(mach, save=False, eta=eta, decay=decay, tol=tol)
def svm(nhidden=30, eta=0.001, decay=0.001, tol=1.25e-4):
templates = data.training_set[0][np.random.randint(data.size_train, size=nhidden),:].T
mach = Machine(data)
mach.add_module(modules.RBFModule, args=[templates])
mach.add_module(modules.SoftMaxModule)
mach.add_module(modules.LinearModule, kwargs={'dim_out': data.nclass})
mach.add_module(modules.BiasModule)
mach.add_module(modules.SoftMaxModule)
mach.add_module(modules.CrossEntropyModule)
run_machine(mach, save=False, eta=eta, decay=decay, tol=tol)
if __name__ == '__main__':
if '--optional' in sys.argv:
data = Dataset('dataset/isolet1+2+3+4.data', test_fn='dataset/isolet5.data',
train=4000, test=1000)
triple_layer(nhidden=40)
rbf_hybrid()
else:
N = 1
if len(sys.argv) > 1:
N = int(sys.argv[1])
for i in range(N):
print "trial %d"%i
data = Dataset('dataset/isolet1+2+3+4.data', test_fn='dataset/isolet5.data',
train=4000, test=1000)
logistic_regression()
single_layer()
double_layer(nhidden=10)
double_layer(nhidden=20)
double_layer(nhidden=40)
double_layer(nhidden=80)