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
| Python 2.7.12 (v2.7.12:d33e0cf91556, Jun 27 2016, 15:24:40) [MSC v.1500 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import theano
WARNING (theano.configdefaults): g++ not available, if using conda: `conda install m2w64-toolchain`
WARNING (theano.configdefaults): g++ not detected ! Theano will be unable to execute optimized C-implementations (for both CPU and GPU) and will default to Python implementations. Performance will be severely degraded. To remove this warning, set Theano flags cxx to an empty string.
>>> import theano.tensor as T
>>> from theano import function, pp, printing, scan, shared
>>> import numpy as np
>>>
>>>
>>> '''Define a function'''
'Define a function'
>>> # 1. theano function
... x, y, a, b = T.fscalars('x', 'y', 'a', 'b')
>>> a = x + y
>>> b = x - y
>>> f_add = theano.function(inputs=[x, y], outputs=a)
>>> f_add_minus = theano.function(inputs=[x, y], outputs=[a, b])
>>>
>>> x1 = 10.
>>> y1 = 1.
>>> a1 = f_add(x1, y1)
>>> a2, b2 = f_add_minus(x1, y1)
>>> print a1, a2, b2
11.0 11.0 9.0
>>>
>>> # 2. python function
... def f_add(x, y):
... a = x + y
... return a
...
>>> def f_add_minus(x, y):
... a = x + y
... b = x - y
... return a, b
...
>>> x1 = 10.
>>> y1 = 1.
>>> a1 = f_add(x1, y1)
>>> a2, b2 = f_add_minus(x1, y1)
>>> print a1, a2, b2
11.0 11.0 9.0
>>>
>>>
>>>
>>> ''' Derivatives '''
' Derivatives '
>>> # Derivative
... x = T.scalar('x', dtype='float32')
>>> y = x ** 2
>>> gy = T.grad(y, x)
>>> f = theano.function([x], gy)
>>> print f(4)
8.0
>>>
>>> # partial derivative
... x = T.scalar('x', dtype='float32')
>>> a = T.scalar('a', dtype='float32')
>>> y = x * a
>>> gy = T.grad(y, [x, a])
>>> f = theano.function([x, a], gy)
>>> print f(4, 3)
[array(3.0, dtype=float32), array(4.0, dtype=float32)]
>>>
>>>
>>> ''' Debugging tricks '''
' Debugging tricks '
>>> x = T.scalar('x',dtype='float32')
>>> y = x ** 2
>>> gy = T.grad(y, x)
>>> f = function([x], gy)
>>>
>>> # using pp
... pp(gy) # before optimization
'((fill((x ** TensorConstant{2}), TensorConstant{1.0}) * TensorConstant{2}) * (x ** (TensorConstant{2} - TensorConstant{1})))'
>>> pp(f.maker.fgraph.outputs[0]) # after optimization
'(TensorConstant{2.0} * x)'
>>>
>>> # using printing.debugprinting
... printing.debugprint(gy) # before optimization
Elemwise{mul} [id A] ''
|Elemwise{mul} [id B] ''
| |Elemwise{second,no_inplace} [id C] ''
| | |Elemwise{pow,no_inplace} [id D] ''
| | | |x [id E]
| | | |TensorConstant{2} [id F]
| | |TensorConstant{1.0} [id G]
| |TensorConstant{2} [id F]
|Elemwise{pow} [id H] ''
|x [id E]
|Elemwise{sub} [id I] ''
|TensorConstant{2} [id F]
|InplaceDimShuffle{} [id J] ''
|TensorConstant{1} [id K]
>>> printing.debugprint(f.maker.fgraph.outputs[0]) # after optimization
Elemwise{mul,no_inplace} [id A] ''
|TensorConstant{2.0} [id B]
|x [id C]
>>>
>>> # printing.Print
... # recall that functions defined before do not print internal variables
... x = T.scalar('x',dtype='float32')
>>> xp2 = x + 2
>>> xp2_printed = printing.Print('this is xp2:')(xp2)
>>> xp2m2 = xp2_printed * 2
>>> f = function([x], xp2m2)
>>> f(20)
this is xp2: __str__ = 22.0
array(44.0, dtype=float32)
>>>
>>> # try to make this work
... x = T.scalar('x',dtype='float32')
>>> y = x ** 2
>>> yprint = printing.Print('this is y:')(y)
>>> gy = T.grad(yprint, x)
>>> f = function([x], gy)
>>> # alternatively:
... # p = printing.Print('y')
... # yprint = p(y)
...
>>>
>>> ''' shared variable '''
' shared variable '
>>> state = shared(0.)
>>> inc = T.iscalar('inc')
>>> accumulator = function([inc], state, updates=[(state, state+inc)])
>>> print state.get_value()
0.0
>>> z = accumulator(10)
>>> print z
0.0
>>> print state.get_value()
10.0
>>>
>>>
>>> ''' Loop '''
' Loop '
>>> # python for loop
... import numpy
>>> A = numpy.array([1, 2], dtype='float32')
>>> k = 5
>>> result = [numpy.array([1,1])]
>>> def mul(a, b): return a*b
...
>>> for i in range(k):
... result.append(mul(result[-1], A))
...
>>> print result[-1]
[ 1. 32.]
>>>
>>> # theano scan
... k = T.scalar("k", dtype='int32')
>>> A = T.vector("A", dtype='float32')
>>> def mul(a, b): return a*b
...
>>> result, updates = theano.scan(
... fn=mul,
... outputs_info=T.ones_like(A),
... non_sequences=A,
... n_steps=k)
>>>
>>> power = theano.function(
... inputs=[A,k],
... outputs=result[-1],
... updates=updates)
>>>
>>> A_val = numpy.array([1, 2], dtype='float32')
>>> k_val = 5
>>> r = power(A_val, 5)
>>> print r
[ 1. 32.]
>>>
>>>
>>> # calculate polynomial
... # that is, a0 * x^0 + a1 * x^1 + a2 * x^2 + ... + an * x^n
... coefficients = theano.tensor.vector("coefficients", dtype='float32')
>>> x = T.scalar("x", dtype='float32')
>>>
>>> max_coefficients_supported = 10000
>>>
>>> def cumulative_poly(coeff, power, prior_sum, x):
... return prior_sum + coeff * (x ** power)
...
>>> # Generate the components of the polynomial
... zero = np.asarray(0., dtype='float32')
>>> full_range=theano.tensor.arange(max_coefficients_supported, dtype='float32')
>>> results, updates = theano.scan(fn=cumulative_poly,
... outputs_info=T.as_tensor_variable(zero),
... sequences=[coefficients, full_range],
... non_sequences=x)
>>>
>>> polynomial = results[-1]
>>> calculate_polynomial = theano.function(inputs=[coefficients, x],
... outputs=polynomial)
>>>
>>> test_coeff = np.asarray([1, 0, 2], dtype=np.float32)
>>> print(calculate_polynomial(test_coeff, 3))
19.0
|
