Kezunlin's Blog

Live and Learn

0%

Example 1-theano linear regression with gradient descent

Posted on Edited on In

Series

code example

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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
#!/usr/bin/python
# -*- coding: utf-8 -*-

# https://blogs.msdn.microsoft.com/lukassteindl/2015/12/13/linear-regression-example-with-python-and-theano/
# http://blog.csdn.net/vins_napoleon/article/details/38057927

from numpy import *
import numpy as np
import theano
import theano.tensor as T
import time

class Linear_Reg(object):
    def __init__(self,x):
        #x,y是scalar,vector(n,),matrix(m,n)
        self.w = theano.shared(value = 0.0,name = 'w')
        self.b = theano.shared(value = 0.0,name = 'b')
        # w,b is scalar,so x can be vector
        self.y_pred = x * self.w + self.b
        self.params = [self.w,self.b]
        
    def msl(self,y):
        return T.sum((y - self.y_pred)**2)

class Linear_Reg2(object):
    def __init__(self,x):
        #x,y是scalar,vector(1,),matrix(m,1)
        self.w = theano.shared(value = np.zeros((1,),dtype=theano.config.floatX),name = 'w')
        self.b = theano.shared(value = np.zeros((1,),dtype=theano.config.floatX),name = 'b')
        # w,b is vector,so x cab be scalar
        self.y_pred = x * self.w + self.b
        self.params = [self.w,self.b]
        
    def msl(self,y):
        return T.sum((y - self.y_pred)**2)

def test_type():
    # x,y be scalar
    points_x = [1.1,2.2] 
    points_y = [3.3,4.4]
    X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX),borrow = True)
    Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX),borrow = True)
    x = T.dscalar('x')
    print x
    print x.type
    print X[0]
    print X[0].type
    
    #x
    #TensorType(float64, scalar)
    #Subtensor{int64}.0
    #TensorType(float64, scalar)
    
def test_type2():
    # x,y be vector
    points_x = [1.1,2.2] 
    points_y = [3.3,4.4]
    X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX),borrow = True)
    Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX),borrow = True)
    x = T.dvector('x')
    print x
    print x.type
    print X[0:2]
    print X[0:2].type
    
    #x
    #TensorType(float64, vector)
    #Subtensor{int64:int64:}.0
    #TensorType(float64, vector)
    

def run_model1(mode):
    """
    [w,b be scalar]
    
    mode = scalar(), set mini_batch_size = 1
    mode = vector(m,),  reshape X,Y to vector  
    mode = matrix(m,n), reshape X,Y to matrix
    """
    eta = 0.000001
    epochs = 1000
    points = genfromtxt("data.csv", delimiter=",")# (100,2)
    points_x = points[:,0] # (100,)   numpy.float64
    points_y = points[:,1] # (100,)   numpy.float64
    N = points_x.shape[0]
    
    if mode == "scalar":
        mini_batch_size = 1  # must be 1 so that all X[i] are used
        X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX),borrow = True)
        Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX),borrow = True)
        
        # so than we get X[0],X[1],...
        x = T.dscalar('tx')
        y = T.dscalar('ty')
    elif mode == "vector":
        mini_batch_size = 5
        X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX),borrow = True)
        Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX),borrow = True)
                      
        # so than we get X[0:5],X[5:10],...
        x = T.dvector('tx')
        y = T.dvector('ty')
    elif mode == "matrix":
        mini_batch_size = 5
        X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX).reshape(N,1),borrow = True)
        Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX).reshape(N,1),borrow = True)
                      
        # so than we get X[0:5],X[5:10],...
        x = T.dmatrix('tx')
        y = T.dmatrix('ty')             
        
    num_batches = N/mini_batch_size
        
    #index = T.lscalar() # l int64
    
    reg = Linear_Reg(x = x)
    cost = reg.msl(y)

    w_g = T.grad(cost = cost, wrt = reg.w)
    b_g = T.grad(cost = cost, wrt = reg.b)

    updates=[(reg.w, reg.w - eta * w_g),
             (reg.b, reg.b - eta * b_g)]

    # use x,y as input (调用函数的时候，x,y的参数化必须是python数值，不能是theano variable)
    train_model = theano.function(inputs=[x,y],
                                  outputs = cost,
                                  updates = updates,
                                  )

    cost_t = 0.0
    costs = []
    start_time = time.clock()

    for epoch in xrange(epochs):
        # 1个epoch,所有N个样本参与训练,mini = m,学习N/m次
        cost_l = []
        for index in range(num_batches):
            if mode == "scalar":
                x = X.get_value()[index]
                y = Y.get_value()[index]
            else:  
                x = X.get_value()[index*mini_batch_size:(index+1)*mini_batch_size]
                y = Y.get_value()[index*mini_batch_size:(index+1)*mini_batch_size] 
            cost_l.append( train_model(x,y) )

        cost_t = np.mean(cost_l)
        costs.append(cost_t)

    end_time = time.clock()
    print '\nTotal time is ：',end_time -start_time,' s' 
    print 'last cost :',cost_t
    print 'w value : ',reg.w.get_value() 
    print 'b value : ',reg.b.get_value() 

    
def run_model2(mode):
    """
    [w,b be vector(1,)]
    
    mode = scalar(),   set mini_batch_size = 1
    mode = vector(1,), set mini_batch_size = 1, reshape X,Y to vector  
    mode = matrix(m,1), reshape X,Y to matrix
    """
    eta = 0.000001
    epochs = 1000
    points = genfromtxt("data.csv", delimiter=",")# (100,2)
    points_x = points[:,0] # (100,)   numpy.float64
    points_y = points[:,1] # (100,)   numpy.float64
    N = points_x.shape[0]
    
    if mode == "scalar":
        mini_batch_size = 1 
        X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX),borrow = True)
        Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX),borrow = True)
        
        # so than we get X[0],X[1],...
        x = T.dscalar('tx')
        y = T.dscalar('ty')
    elif mode == "vector":
        mini_batch_size = 1 
        X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX),borrow = True)
        Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX),borrow = True)
                      
        # so than we get X[0:5],X[5:10],...
        x = T.dvector('tx')
        y = T.dvector('ty')
    elif mode == "matrix":
        mini_batch_size = 5
        X = theano.shared(np.asarray(points_x,dtype=theano.config.floatX).reshape(N,1),borrow = True)
        Y = theano.shared(np.asarray(points_y,dtype=theano.config.floatX).reshape(N,1),borrow = True)
                      
        # so than we get X[0:5],X[5:10],...
        x = T.dmatrix('tx')
        y = T.dmatrix('ty')             
        
    num_batches = N/mini_batch_size
        
    index = T.lscalar() # l int64
    reg = Linear_Reg2(x = x)
    cost = reg.msl(y)

    #w_g,b_g = T.grad(cost,[reg.w,reg.b])
    #updates=[(reg.w, reg.w - eta * w_g), (reg.b, reg.b - eta * b_g)]

    #==========================================================================
    # use params and updates
    #==========================================================================
    params = [reg.w,reg.b]       # list of [w,b]
    grads = T.grad(cost,params)  # list of [w_g,b_g]
    updates = [(param,param-eta*grad) 
               for param,grad in zip(params,grads)]               
                                # list of [ (w,w-eta*w_g), (b, b-eta*b_g) ]
        
    # updates必须是shared变量;
    # use x,y as input (调用函数的时候，x,y的参数化必须是python数值，不能是theano variable)
    # use index as input (调用函数的时候，index的参数化必须是python数值),
    # 通过givens替换掉x,y，X[index],Y[index]和x,y的类型必须一致，都是theano variable
    # 此处x,y，X[index],Y[index]都是  TensorType(float64, scalar)　
    # 或者是float64, vector,matrix类型
    if mode == "scalar":
        train_model = theano.function(inputs=[index],
                                      outputs = cost,
                                      updates = updates,
                                      givens = {x:X[index],
                                                y:Y[index]})
    else:
        train_model = theano.function(inputs=[index],
                                      outputs = cost,
                                      updates = updates,
                                      givens = {x:X[index*mini_batch_size:(index+1)*mini_batch_size],
                                                y:Y[index*mini_batch_size:(index+1)*mini_batch_size]})

    cost_t = 0.0
    costs = []
    start_time = time.clock()

    for epoch in xrange(epochs):
        # 1个epoch,所有N个样本参与训练,mini = m,学习N/m次
        cost_l = []
        for index in range(num_batches):
            cost_l.append( train_model(index) )
            
        cost_t = np.mean(cost_l)
        costs.append(cost_t)

    end_time = time.clock()
    print '\nTotal time is ：',end_time -start_time,' s' 
    print 'last cost :',cost_t
    print 'w value : ',reg.w.get_value()  
    print 'b value : ',reg.b.get_value()


"""
run:
Total time is ： 2.796644  s
last cost : 113.178407322
w value :  1.48497718432
b value :  0.0890071567283

run2:
Total time is ： 2.127487  s
last cost : 113.178407322
w value :  [ 1.48497718]
b value :  [ 0.08900716]

run3:
Total time is ： 0.480144  s
avg cost : 565.419900755
w value :  [ 1.48492605]
b value :  [ 0.08896923]
"""
  
def test():
    test_type()
    test_type2()
    
if __name__ == '__main__':
    #test()
    
    #run_model1("scalar")
    #run_model1("vector")   
    #run_model1("matrix")   
    
    run_model2("scalar")
    run_model2("vector")   
    run_model2("matrix") 
Total time is ： 1.929041  s
last cost : 113.178407322
w value :  [ 1.48497718]
b value :  [ 0.08900716]

Total time is ： 2.040197  s
last cost : 113.178407322
w value :  [ 1.48497718]
b value :  [ 0.08900716]

Total time is ： 0.553913  s
last cost : 565.419900755
w value :  [ 1.48492605]
b value :  [ 0.08896923]

Reference

History

  • 20180807: created.