-
Notifications
You must be signed in to change notification settings - Fork 0
/
EKF_LSTM.py
192 lines (140 loc) · 5.91 KB
/
EKF_LSTM.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
192
# -*- coding: utf-8 -*-
from collections import OrderedDict
import numpy
import theano
import theano.tensor as T
import time
import sys
import matplotlib.pyplot as plt
import copy
import utilities.datagenerator as DG
reload(DG)
compile_mode = 'FAST_COMPILE'
# compile_mode = 'FAST_RUN'
# Set the random number generators' seeds for consistency
# SEED = int(numpy.random.lognormal()*100)
SEED = 123
numpy.random.seed(SEED)
def lstm_layer(n_input, n_LSTM, x):
'''
i f o c 统一处理
'''
def _slice(_x, n, dim):
if _x.ndim == 3:
return _x[:, :, n * dim:(n + 1) * dim]
return _x[:, n * dim:(n + 1) * dim]
def _step(*args):
'''
x_ : 延时输入的x
h_ : 前一时刻单元的输出
c_ : 前一时刻单元的Cell值
'''
x = [args[u] for u in xrange(n_input)]
h_ = args[n_input]
c_ = args[n_input+1]
preact = T.dot(x[0], W_in[0])
for i in xrange(1,n_input):
preact += T.dot(x[i], W_in[i])
preact += T.dot(h_, W_hid) # h的后向
preact += b_in
i = T.nnet.sigmoid(_slice(preact, 0, n_LSTM)) # input gate
f = T.nnet.sigmoid(_slice(preact, 1, n_LSTM)) # forget gate
o = T.nnet.sigmoid(_slice(preact, 2, n_LSTM)) # output gate
c = T.tanh(_slice(preact, 3, n_LSTM)) # cell state pre
c = f * c_ + i * c # cell state
h = o * T.tanh(c) # unit output
return h, c
out_h = theano.shared(numpy.zeros((1,n_LSTM), dtype=theano.config.floatX), name="out_h")
out_c = theano.shared(numpy.zeros((1,n_LSTM), dtype=theano.config.floatX), name="out_c")
input_taps = range(1-n_input, 1)
rval, updates = theano.scan(_step,
sequences=dict(input=x,taps=input_taps),
outputs_info=[out_h, out_c])
return rval[0] # 对外只有h
build_method = 5 # 0: LSTM
init_method = 0 # 0: normal 1: uniform
# 设置网络参数
n_input = 7
n_hidden = 15
n_output = 1
n_epochs = 5
saveto = 'MaskRNN_b{}_i{}_h{}_nh{}_S{}.npz'.format(
build_method, init_method, n_hidden, 0, SEED)
print 'Result will be saved to: ', saveto
dtype=theano.config.floatX
theano.config.exception_verbosity = 'low'
# 加要处理的数据
g = DG.Generator()
data_x,data_y = g.get_data('mackey_glass')
index_test_begin = data_y.shape[0] / 2
train_data_index = numpy.arange(index_test_begin)
test_data_index = numpy.arange(index_test_begin, data_y.shape[0])
train_data = data_y[train_data_index]
test_data = data_y[test_data_index]
print 'train_data.shape: ', train_data.shape
print 'test_data.shape: ', test_data.shape
print 'network: n_in:{},n_hidden:{},n_out:{}'.format(n_input, n_hidden, n_output)
x = T.vector() # 输入向量,第1维是时间
y = T.vector() # 输出向量, 第1维是时间
W_in = [theano.shared(numpy.random.uniform(size=(1, 4*n_hidden), low= -0.01, high=0.01).astype(dtype),
name='W_in' + str(u)) for u in range(n_input)]
b_in_value = numpy.zeros((4 * n_hidden,), dtype=dtype)
b_in_value[1*n_hidden : 2*n_hidden] = 5 # large forget gate
b_in = theano.shared(numpy.zeros((4 * n_hidden,), dtype=dtype), name="b_in")
W_hid = theano.shared(numpy.random.uniform(size=(n_hidden, 4*n_hidden), low= -0.01, high=0.01).astype(dtype), name='W_hid')
W_out = theano.shared(numpy.random.uniform(size=(n_hidden,n_output),low=-0.01,high=0.01).astype(dtype),name="W_out")
b_out = theano.shared(numpy.zeros((n_output,), dtype=dtype),name="b_out")
params = []
params.extend(W_in)
params.extend([b_in])
params.extend([W_hid])
params.extend([W_out])
params.extend([b_out])
h_tmp = lstm_layer(n_input, n_hidden, x)
pred = T.dot(h_tmp, W_out) + b_out
pred = theano.tensor.flatten(pred)
f_pred = theano.function([x], pred, name='f_pred')
cost = ((pred - y)**2).sum()
batch_size = 2 # 设置的足够大时，等价于GD
print 'Batch Size: ', batch_size
update_W, P, cost = DG.PublicFunction.extend_kalman_train(params, pred, batch_size, y)
f_train = theano.function([x, y], cost, updates=update_W,
name='EKF_f_train',
mode=compile_mode)
sim_fn = theano.function([x],outputs=pred)
start_time = time.clock()
for epochs_index in xrange(n_epochs) :
kf = DG.DataPrepare.get_seq_minibatches_idx(train_data.shape[0], batch_size, n_input, shuffle=False)
for batch_index, train_index in kf:
sub_seq = train_data[train_index]
_x, _y = DG.PublicFunction.data_get_data_x_y(sub_seq, n_input)
train_err = f_train(_x, _y)
print '{}.{}: cost={}'.format(epochs_index, batch_index, train_err)
x_train_end = copy.deepcopy(train_data[-n_input:])
n_predict = 150
y_predict = numpy.zeros((n_predict,))
cumulative_error = 0
cumulative_error_list = numpy.zeros((n_predict,))
for i in numpy.arange(n_predict):
y_predict[i] = sim_fn(x_train_end)
x_train_end[:-1] = x_train_end[1:]
x_train_end[-1] = y_predict[i]
cumulative_error += numpy.abs(y_predict[i] - test_data[i])
cumulative_error_list[i] = cumulative_error
plt.figure(3)
plt.plot(numpy.arange(n_predict), cumulative_error_list)
plt.title('cumulative error')
plt.grid(True)
plt.figure(1)
plt.plot(numpy.arange(y_predict.shape[0]), y_predict,'r')
plt.plot(numpy.arange(300), test_data[:300],'g')
y_sim = sim_fn(data_x[:-1]) # 整体的单步误差
print 'y_sim.shape: ', y_sim.shape
plt.figure(2)
plt.plot(range(data_y.shape[0]), data_y,'k')
plt.plot(range(data_y.shape[0]-y_sim.shape[0], data_y.shape[0]), y_sim, 'r')
plt.plot(range(data_y.shape[0]-y_sim.shape[0], data_y.shape[0]), y_sim - data_y[n_input:], 'g')
print >> sys.stderr, ('overall time (%.5fs)' % ((time.clock() - start_time) / 1.))
numpy.savez(saveto, cumulative_error=cumulative_error_list)
plt.show()
print "finished!"