tensorflow-101/python/DNNRegressor.py at master · Tantael/tensorflow-101 · GitHub

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
import tensorflow as tf
import numpy as np

from tensorflow.contrib.learn.python.learn.utils import input_fn_utils #export_savedmodel

import matplotlib.pyplot as plt
import io

#----------------------------

input = np.loadtxt("sine.csv", dtype='f', delimiter=',')
#print(input)

row = input.shape[0]
col = input.shape[1]

#----------------------------
#attributes and labels
attributes = [[0 for i in range(col-1)] for j in range(row)]
labels = []
for i in range(row):
	labels.append(0)

for i in range(0, row):
	for j in range(0, col):
		if j < col -1:
			attributes[i][j] = input[i][j]
		else:
			labels[i] = input[i][j]

data =np.array(attributes, 'float32')
target = np.array(labels, 'float32')

#----------------------------

#nn learning parameters
learningRate = 0.1
epoch = 10000

#----------------------------

feature_columns = [tf.contrib.layers.real_valued_column("", dimension = col-1)]

#neural network model
regressor = tf.contrib.learn.DNNRegressor(
	feature_columns = feature_columns
	, hidden_units = [4] #a hidden layer consisting of 4 nodes
	, optimizer = tf.train.GradientDescentOptimizer(learningRate)
	, activation_fn = tf.nn.sigmoid
	, model_dir = "model" #model will be stored in this folder
	, config = tf.contrib.learn.RunConfig(save_checkpoints_secs = 1)
)

validation_monitor = tf.contrib.learn.monitors.ValidationMonitor(data, target, every_n_steps = 1000)

regressor.fit(data, target, steps = epoch
	, monitors = [validation_monitor]
)

"""
#this block provides to export nn model language neutrally. in this way, same model can be used in high level systems such as Java
feature_spec = tf.contrib.layers.create_feature_spec_for_parsing(feature_columns)
serving_input_fn = input_fn_utils.build_parsing_serving_input_fn(feature_spec)
regressor.export_savedmodel(regressor.model_dir+"\export", serving_input_fn, as_text=True)
"""

def test_set():
	return np.array(attributes, np.float32)

predictions = regressor.predict_scores(input_fn = test_set)

#--------------------------
#dumping predictions and actual sets

index = 0
for i in predictions:
	print("actual: ", target[index],", predic: ", i)
	index = index + 1

#--------------------------

#model performance
eva = regressor.evaluate(data, target)
print("MSE: ", eva["loss"])

#--------------------------

#visualizing predictions and actuals in TensorBoard

#predictions have to be restored for processing
predictions = regressor.predict_scores(input_fn = test_set)

actuals = labels
forecasts = list(predictions)

forecast_writer = tf.summary.FileWriter('model/forecast')
actual_writer = tf.summary.FileWriter('model/actual')

for i in range(0, row):
	actual_summary = tf.Summary(
		value = [tf.Summary.Value(tag="summary_tag", simple_value= actuals[i])])
	forecast_summary = tf.Summary(
		value = [tf.Summary.Value(tag="summary_tag", simple_value= forecasts[i])])

	actual_writer.add_summary(actual_summary, i)
	forecast_writer.add_summary(forecast_summary, i)

#--------------------------

"""
#x-axis: time, y-axis:value. actual and forecast values are plotted as 2 line in same graph.
#this graph cannot be readable for displaying too many point
plt.plot(actuals)
plt.plot(forecasts)
plt.show()
"""

"""
#x-axis represents predicted values whereas y-axis represents actual values.
plt.scatter(actuals, forecasts)
plt.xlabel('predicted')
plt.ylabel('actual')
plt.show()
"""

#----------------------------------
#embedding matplotlib graph in TensorBoard

def create_plot(actuals, forecasts):
	plt.figure()
	plt.scatter(actuals, forecasts)
	buf = io.BytesIO()
	plt.savefig(buf, format='png')
	buf.seek(0)
	return buf

plot_buf = create_plot(actuals, forecasts)
image = tf.image.decode_png(plot_buf.getvalue(), channels=4)
image = tf.expand_dims(image, 0)

summary_img = tf.summary.image("scatter_plot", image)

sess = tf.Session()

summary = sess.run(summary_img)
writer = tf.summary.FileWriter('model/logs')
writer.add_summary(summary)