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
|
#!/usr/bin/python3
from sklearn import datasets
from sklearn import preprocessing
from sklearn.cross_validation import train_test_split
import numpy as np
import tensorflow as tf
dataset = datasets.load_boston()
data, target = dataset.data, dataset.target
data_scaler = preprocessing.MinMaxScaler()
data = data_scaler.fit_transform(data)
np.random.seed(0)
x_train, x_test, y_train, y_test = train_test_split(
data, target, train_size=0.9
)
with tf.device('/cpu:0'):
sess = tf.InteractiveSession()
input_matrix = tf.placeholder(tf.float32, [None, 13], name="input")
real = tf.placeholder(tf.float32, [1, None], name="target")
fact = tf.placeholder(tf.float32, [1, None], name="factor")
# ^ max(input_matrix) - min(input_matrix)
offs = tf.placeholder(tf.float32, [1, None], name="offset")
# ^ min(input_matrix)
fitted_real = tf.div(tf.sub(real, offs), fact) # input scaled
weights = tf.Variable(tf.truncated_normal([13], stddev=0.1), name="weight")
bias = tf.Variable(0.0, name="bias")
weighted = tf.reduce_sum(tf.mul(input_matrix, weights), 1)
# ^ sum ([13]weights*input_matrix)
fitted_prediction = tf.sigmoid(tf.add(weighted, bias))
prediction = tf.add(tf.mul(fitted_prediction, fact), offs)
# ^ un-scale prediction
diff = tf.sub(prediction, real) # for testing
sqr = tf.square(diff)
mean = tf.reduce_mean(sqr)
mse = mean
difference = tf.abs(
tf.reduce_sum(tf.sub(fitted_prediction, fitted_real)))
mse_summary = tf.scalar_summary("mse", mse)
diff_summary = tf.scalar_summary("max diff", tf.reduce_max(diff))
diff_summary = tf.scalar_summary("mean diff", tf.reduce_mean(diff))
merged = tf.merge_all_summaries()
writer = tf.train.SummaryWriter("/tmp/boston_logs", sess.graph_def)
# all variables have to be specified here
sess.run(tf.initialize_all_variables())
mse = tf.Print(mse, [mse], "mse: ")
train_step = tf.train.GradientDescentOptimizer(0.0001).minimize(mse)
last = 0
batchsize = 100
next_last = last + batchsize
factor = [[max(target) - min(target)]]
offset = [[min(target)]]
while next_last < len(x_train):
print("running next batch")
next_last = last + batchsize
if next_last > len(x_train):
next_last = len(x_train)
xt = x_train[last:next_last]
yt = [y_train[last:next_last]]
for i in range(0, 100): # 100 epochs
if i % 100 == 0:
feed = {input_matrix: x_test, real: [y_test], fact: factor, offs: offset}
result = sess.run([merged, mse, difference], feed_dict=feed)
writer.add_summary(result[0], i)
sess.run(train_step,
feed_dict={input_matrix: xt, real: yt, fact: factor, offs: offset})
last = next_last
print("finished training")
yt = [y_test]
# debug
print(" --------- ")
print("mean difference to test data: ")
print(sess.run(mse, feed_dict={input_matrix: x_test, real: yt, fact: factor, offs: offset}))
print(sess.run(diff, feed_dict={input_matrix: x_test, real: yt, fact: factor, offs: offset}))
|
