TensorFlow-API

Posted on Edited on In Symbols count in article: 22k Reading time ≈ 20 mins.
TensorFlow常用API

tf.constant

根据value的值，生成一个shape维度的张量常量

Define

1
constant(value, dtype=None, shape=None, name='Const', verify_shape=False)

Parameter

参数名 必须 类型 说明
value Y 常量、list 张量的值
dtype N dtype 张量的元素类型
shape N 1维整型张量、array 张量的维度
name N string 张量的名称
verify_shape N boolean 检测shape是否和value的shape一致。
若值为False，当不一致时，会用最后一个元素将shape补全

Demo

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
import tensorflow as tf
import numpy as np

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: constant.py
@time: 2018/5/11 10:00
"""

a = tf.constant([1,2,3,4,5,6], shape=[2,3])
b = tf.constant(-1, shape=[3,2])
c = tf.matmul(a,b)

e = tf.constant(np.arange(1,13,dtype=np.int32), shape=[2,2,3])
f = tf.constant(np.arange(13,25, dtype=np.int32), shape=[2,3,2])
g = tf.matmul(e,f)

with tf.Session() as sess:
    print ("-----------------")
    print (a.get_shape())
    print (sess.run(a))
    print ("-----------------")
    print (b.get_shape())
    print (sess.run(b))
    print ("-----------------")
    print (c.get_shape())
    print (sess.run(c))
    print ("-----------------")
    print (e.get_shape())
    print (sess.run(e))
    print ("-----------------")
    print (f.get_shape())
    print (sess.run(f))
    print ("-----------------")
    print (g.get_shape())
    print (sess.run(g))
    
# 执行结果
# -----------------
# (2, 3)
# [[1 2 3]
#  [4 5 6]]
# -----------------
# (3, 2)
# [[-1 -1]
#  [-1 -1]
#  [-1 -1]]
# -----------------
# (2, 2)
# [[ -6  -6]
#  [-15 -15]]
# -----------------
# (2, 2, 3)
# [[[ 1  2  3]
#   [ 4  5  6]]
#
#  [[ 7  8  9]
#   [10 11 12]]]
# -----------------
# (2, 3, 2)
# [[[13 14]
#   [15 16]
#   [17 18]]
#
#  [[19 20]
#   [21 22]
#   [23 24]]]
# -----------------
# (2, 2, 2)
# [[[ 94 100]
#   [229 244]]
#
#  [[508 532]
#   [697 730]]]

tf.placeholder

一种占位符，在执行时填入具体值

###Define

1
placeholder(dtype, shape=None, name=None)

Parameter

参数名 必选 类型 说明
dtype Y dtype 占位符的数据类型
shape N 1维整型张量、array 占位符的维度
name N string 占位符的名称

Demo

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
import tensorflow as tf
import numpy as np

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: placeholder.py
@time: 2018/5/11 10:57
"""

x = tf.placeholder(tf.float32, [None, 3])
y = tf.matmul(x, x)
with tf.Session() as sess:
    rand_array = np.random.rand(3,3)
    print (sess.run(y, feed_dict={x:rand_array}))

# 执行结果
# [[ 0.88505751  0.79640609  0.46087515]
#  [ 0.36539319  0.41709256  0.16832483]
#  [ 0.75206953  0.90277183  0.37584797]]

tf.nn.conv2d

卷积运算

Define

1
2
conv2d(input, filter, strides, padding, 
		use_cudnn_on_gpu=True, data_format='NHWC', name=None)

Parameter

参数名 必选 类型 说明
input Y tensor 4维的tensor，即[batch, in_height, in_width, in_channels]
(若input是图像，即[训练时一个batch的图片数量，图片高度，图片宽度，图片的通道数])
类型需为float32、float64
filter Y tensor 4维的tensor，即[filter_height, filter_width, in_channels, out_channels]
(若input是图像，即[卷积核的高度，卷积核的宽度，图像通道数，卷积核个数])
filter的in_channels必须与input的in_channels相等
strides Y list 长度为4的list，卷积时在input上每一维的步长，一般strides[0]=strides[3]=1
padding Y string 卷积方式，只能为”SAME”、”VALID”其中之一。SAME：补全方式；VALID：丢弃方式
use_cudnn_on_gpu N boolean 是否使用cudnn加速，默认为true
data_format N string 只能为”NHWC”、”NCHW”，默认”NHWC”
Name N string 运算名称

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: conv2d.py
@time: 2018/5/9 20:48
"""

a = tf.constant([1,1,1,0,0,0,1,1,1,0,0,0,1,1,1,0,0,1,1,0,0,1,1,0,0],dtype=tf.float32,shape=[1,5,5,1])
b = tf.constant([1,0,1,0,1,0,1,0,1],dtype=tf.float32,shape=[3,3,1,1])
c = tf.nn.conv2d(a,b,strides=[1, 2, 2, 1],padding='VALID')
d = tf.nn.conv2d(a,b,strides=[1, 2, 2, 1],padding='SAME')
with tf.Session() as sess:
    print ("c shape:")
    print (c.get_shape())
    print ("c value:")
    print (sess.run(c))
    print ("d shape:")
    print (d.get_shape())
    print ("d value:")
    print (sess.run(d))

# 运算结果
# c shape:
# (1, 2, 2, 1)
# c value:
# [[[[ 4.]
#    [ 4.]]
#
#   [[ 2.]
#    [ 4.]]]]
# d shape:
# (1, 3, 3, 1)
# d value:
# [[[[ 2.]
#    [ 3.]
#    [ 1.]]
#
#   [[ 1.]
#    [ 4.]
#    [ 3.]]
#
#   [[ 0.]
#    [ 2.]
#    [ 1.]]]]

tf.nn.relu

激活函数

Define

1
relu(features, name=None)

Parameter

参数名 必选 类型 说明
features Y tensor float32、float64、int8、int16、int32、int64、uint8、unit16、half
name N string 运算名称

Demo

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: relu.py
@time: 2018/5/9 21:30
"""

a = tf.constant([1,-2,0,4,-5,6])
b = tf.nn.relu(a)
with tf.Session() as sess:
    print (sess.run(b))

# 运算结果
# [1 0 0 4 0 6]

tf.nn.max_pool

池化

Define

1
2
max_pool(value, ksize, strides, padding,
		data_format='NHWC', name=None)

Parameter

参数名 必选 类型 说明
value Y tensor 4维的张量，即[batch, height, width, channels]，类型为tf.float32
ksize Y list 池化窗口的大小，长度为4的list，一般是[1，height，width，1]，因为不在batch和channels上做池化，所以第一个和最后一个维度为1
strides Y list 池化窗口在每一个维度上的步长。一般strides[0]=strides[3]=1
padding Y string 只能为”VALID”、”SAME”其中之一，这个值决定了不同的池化方式。VALID：丢弃方式；SAME：补全方式
data_format N string 只能为”NHWC”、”NCHW”。默认”NHWC”
name N string 运算名称

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: max_pool.py
@time: 2018/5/10 20:38
"""

a = tf.constant([1,3,2,1,2,9,1,1,1,3,2,3,5,6,1,2], dtype=tf.float32, shape=[1,4,4,1])
b = tf.nn.max_pool(a, ksize=[1,2,2,1], strides=[1, 2, 2, 1], padding='SAME')
c = tf.nn.max_pool(a, ksize=[1,2,2,1], strides=[1,2,2,1], padding='VALID')
with tf.Session() as sess:
    print ("b.shape")
    print (b.get_shape())
    print ("b.value")
    print (sess.run(b))

    print ("c.shape")
    print (c.get_shape())
    print ("c.value")
    print (sess.run(c))

# 运算结果
# b.shape
# (1, 2, 2, 1)
# b.value
# [[[[ 9.]
#    [ 2.]]
#
#   [[ 6.]
#    [ 3.]]]]
# c.shape
# (1, 2, 2, 1)
# c.value
# [[[[ 9.]
#    [ 2.]]
#
#   [[ 6.]
#    [ 3.]]]]

tf.nn.dropout

丢弃函数。在训练过程中，按照一定概率将一部分神经网络单元暂时从网络中丢弃，相当于从原始网络中找到一个更瘦的网络

Define

1
dropout(x, keep_prob, noise_shape=None, seed=None, name=None)

Parameter

参数名 必选 类型 说明
x Y tensor x中keep_prob概率的元素，除以keep_prob。或者为0
keep_prob Y scalar dropout的概率，一般先用占位符
noise_shape N tensor 默认情况下，每个元素是否dropout是相互独立的。
如果noise_shape[i]==shape(x)[i]，该维度元素相互独立；
如果noise_shape[i]!=shape(x)[i]，该维度元素不相互独立，即要么一起dropout，要么一起保留
seed N 数值 如果指定该值，每次dropout结果相同
name N string 运算名称

Demo

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: dropout.py
@time: 2018/5/10 20:58
"""

a = tf.constant([1,2,3,4,5,6], shape=[2,3], dtype=tf.float32)
b = tf.placeholder(tf.float32)
c = tf.nn.dropout(a, b, [2,1], 1)
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print (sess.run(c, feed_dict={b:0.75}))

# 运算结果
# [[ 0.          0.          0.        ]
#  [ 5.33333349  6.66666651  8.        ]]

tf.nn.sigmoid_cross_entropy_with_logits

计算交叉熵函数。先对 logits 进行 sigmoid 计算，再计算交叉熵

交叉熵：Loss函数的一种。常见的Loss函数为均方平方差；

sigmoid：激活函数的一种；

另外除了tf.nn.sigmoid_cross_entropy_with_logits以外，还有

tf.nn.softmax_cross_entropy_with_logits（参数logits必须为未经缩放的，函数中会对logits进行sigmoid运算）

tf.nn.sparse_softmax_cross_entropy_with_logits

tf.nn.weighted_cross_entropy_with_logits三种交叉熵函数

Define

1
sigmoid_cross_entropy_with_logits(_sentinel=None, labels=None, logits=None, name=None)

Parameter

参数名 必选 类型 说明
_sentinel N None 一般不使用
logits N tensor float32、float64
labels N tensor shape与logits相同
name N string 运算名称

Demo

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: sigmod_cross_entropy_with_logits.py.py
@time: 2018/5/11 09:27
"""

x = tf.constant([1,2,3,4,5,6,7], dtype=tf.float64)
y = tf.constant([1,1,1,0,0,1,0], dtype=tf.float64)
loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=x, labels=y)

with tf.Session() as sess:
    print (sess.run(loss))

# 运算结果
# [3.13261688e-01   1.26928011e-01   4.85873516e-02   4.01814993e+00
#  5.00671535e+00   2.47568514e-03   7.00091147e+00]

该函数返回值是一个向量，并不是一个数。

如果要求交叉熵，需要进一步做tf.reduce_sum操作，即对向量里所有元素求和；

如果要求Loss（常见），需要进一步做tf.reduce_mean操作，对向量求平均值；

tf.nn.bias_add

将偏差项 bias 加到 value 上面，可以看做是 tf.add 的一个特例。其中 bias 必须是一维的，并且维度和 value 的最后一维相同，数据类型必须和 value 相同

Define

1
bias_add(value, bias, data_format=None, name=None)

Parameter

参数名 必选 类型 说明
value Y tensor Float、double、int64、int32、int16、int8、uint8、complex64、complex128
bias Y 1维tensor 维度必须和value最后一维维度相等
data_format N string 支持”NHWC”、”NCHW”
name N string 运算名称

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: bias_add.py
@time: 2018/5/11 11:45
"""

a = tf.constant([[1.0, 2.0],[1.0, 2.0],[1.0, 2.0]])
b = tf.constant([2.0,1.0])
c = tf.constant([1.0])
with tf.Session() as sess:
    print ("bias_add(a, b)")
    print (sess.run(tf.nn.bias_add(a, b)))
    # print (sess.run(tf.nn.bias_add(a, c))) error
    print ("-----------------")
    print ("tf.add(a, b)")
    print (sess.run(tf.add(a, b)))
    print ("-----------------")
    print ("tf.add(a, c")
    print (sess.run(tf.add(a, c)))


# bias_add(a, b)
# [[ 3.  3.]
#  [ 3.  3.]
#  [ 3.  3.]]
# -----------------
# tf.add(a, b)
# [[ 3.  3.]
#  [ 3.  3.]
#  [ 3.  3.]]
# -----------------
# tf.add(a, c
# [[ 2.  3.]
#  [ 2.  3.]
#  [ 2.  3.]]

tf.truncated_normal

在截断的正态分布中输出随机数，取值范围[ mean - 2 * stddev, mean + 2 * stddev ]

Define

1
truncated_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)

Parameter

参数名 必选 类型 说明
shape Y 1维整型张量、array 输出张量的维度
mean N 0维张量、数值 均值
stddev N 0维张量、数值 标准差
dtype N dtype 输出类型
seed N 数值 随机种子，若seed复制，每次产生相同随机数
name N string 运算名称

Demo

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: truncated_nomal.py
@time: 2018/5/11 09:52
"""

initial = tf.truncated_normal(shape=[3,3], mean=0, stddev=1)
print (tf.Session().run(initial))

# 运算结果
# [[-0.49039015  0.67874366  1.00580561]
#  [ 0.97545707  0.15845388 -0.73228759]
#  [-0.60446167 -0.87802863  0.24452373]]

tf.reduce_mean

计算张量input_tensor的平均值

Define

1
reduce_mean(input_tensor, axis=None, keep_dims=False, name=None, reduction_indices=None)

Parameter

参数名 必选 类型 说明
input_tensor Y 张量 待求平均值的张量
axis N None、0、1 None：全局球平均值
0：求每一列平均值
1：求每一行平均值
keep_dims N boolean 保留原来的维度（例如不会从二维矩阵降为一维）
name N string 运算名称
reduction_indices N None 和axis等价，已弃用

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: reduce_mean.py
@time: 2018/5/11 13:46
"""

initial = [[1.,1.],[2.,2.]]
x = tf.Variable(initial, dtype=tf.float32)
init_op = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init_op)
    print (sess.run(tf.reduce_mean(x)))
    print (sess.run(tf.reduce_mean(x, 0)))  #column
    print (sess.run(tf.reduce_mean(x, 1)))  #row

# 运算结果
# 1.5
# [ 1.5  1.5]
# [ 1.  2.]

tf.square

计算张量对应元素的平方

Define

1
square(x, name=None)

Parameter

参数名 必选 类型 说明
x Y tensor half、float32、float64、int32、int64、complex64、complex128
name N string 运算名称

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: square.py
@time: 2018/5/11 13:59
"""

initial_x = [[1.,1.], [2.,2.]]
x = tf.Variable(initial_x, dtype=tf.float32)
x_2 = tf.square(initial_x)

init_op = tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init_op)
    print (sess.run(x_2))

# 运算结果
# [[ 1.  1.]
#  [ 4.  4.]]

tf.squared_difference

计算张量x、y对应元素差平方

Define

1
squared_difference(x, y, name=None)

Parameter

参数名 必选 类型 说明
x Y tensor half、float32、float64、int32、int64、complex64、complex128
y Y tensor half、float32、float64、int32、int64、complex64、complex128
name N string 运算名称

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: squared_difference.py
@time: 2018/5/11 13:54
"""

initial_x = [[1.,1.], [2.,2.]]
x = tf.Variable(initial_x, dtype=tf.float32)
initial_y = [[3.,3.], [4.,4.]]
y = tf.Variable(initial_y, dtype=tf.float32)

diff = tf.squared_difference(x, y)
init_op = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init_op)
    print (sess.run(diff))

# 运算名称
# [[ 4.  4.]
#  [ 4.  4.]]

tf.Variable

类，维护图在执行过程中的状态信息，例如神经网络权重值的变化

Define

1
2
3
__init__(initial_value=None, trainable=True, collections=None, validate_shape=True,
		caching_device=None, name=None, variable_def=None, dtype=None, expected_shape=None,
		import_scope=None)

Parameter

参数名 类型 说明
initial_value tensor Variable类的初始值。这个张量必须制定shape信息，否则后面validate_shape需设为False
trainable boolean 是否把变量添加到 GraphKeys.TRAINABLE_VARIABLES中。
（ collection 是一种全局存储，不受变量名生存空间影响，一处保存，到处可取）
collections Graph collections 全局存储，默认是 GraphKeys.GLOBAL_VARIABLES
validate_shape boolean 是否允许被位置维度的initial_value初始化
caching_device string 指明哪个device用来缓存变量
name string 变量名
dtype dtype 如果设置，初始化的值按照该类型初始化
expected_shape tensorShape 要是设置了，那么初始的值会是这种维度

Demo

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
import tensorflow as tf

"""
@version: ??
@author: Gavin
@license: Apache Licence 
@contact: woniuyf@gmail.com
@site: https://wonius.top/
@software: PyCharm
@file: Variable.py
@time: 2018/5/11 14:46
"""


initial = tf.truncated_normal(shape=[10, 10], mean=0, stddev=1)
W = tf.Variable(initial)

list = [[1.,1.], [2.,2.]]
X = tf.Variable(list, dtype=tf.float32)

init_op = tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init_op)
    print ("------------------------------------------------------------")
    print (sess.run(W))

    print ("------------------------------------------------------------")
    print (sess.run(W[:2,:2]))
    op = W[:2,:2].assign(22.*tf.ones((2,2)))

    print ("------------------------------------------------------------")
    print (sess.run(op))

    print ("------------------------------------------------------------")
    print (W.eval(sess)) # computes and returns the value of this variable
    print (W.eval())    # user with the default session

    print ("------------------------------------------------------------")
    print (W.dtype)
    print (sess.run(W.initial_value))
    print (sess.run(W.op))
    print (W.shape)

    print ("------------------------------------------------------------")
    print (sess.run(X))

# 运算结果
# ------------------------------------------------------------
# [[ -1.36460924e+00  -8.78410161e-01   1.87626636e+00   1.27253577e-01
#     1.77748889e-01  -2.79777735e-01  -1.29324961e+00   1.18464601e+00
#     5.91014000e-03   3.91639054e-01]
#  [  1.09924674e+00  -1.08868659e+00  -1.12289786e+00  -6.78496242e-01
#     4.25694019e-01   1.06920172e-02   8.05350840e-01  -7.63205171e-01
#     3.23375136e-01   4.49576497e-01]
#  [ -1.05841234e-01  -2.85442650e-01   1.01552248e+00  -1.69130695e+00
#    -1.15725782e-03   1.47837210e+00   6.69513494e-02   4.84152675e-01
#    -3.82478178e-01  -1.20722222e+00]
#  [  3.09620090e-02  -7.98057988e-02   5.60789108e-01  -1.78828037e+00
#     1.95968688e-01  -2.27404922e-01  -1.72476351e+00  -1.23355138e+00
#    -9.63081241e-01  -1.60457373e+00]
#  [  3.23699638e-02  -5.90677261e-01  -1.92470342e-01  -4.10615206e-01
#     3.35628651e-02  -1.41288012e-01  -3.40104014e-01   1.13289797e+00
#     6.52421772e-01   9.17639136e-01]
#  [  5.11528432e-01   1.64260656e-01  -7.84879684e-01  -2.26418376e-01
#    -8.90385628e-01  -3.38358462e-01   1.96581781e+00  -7.17314780e-02
#     1.23814851e-01   7.64766455e-01]
#  [  2.22003132e-01  -1.70102561e+00  -1.60155982e-01   1.10927892e+00
#    -1.14349496e+00  -6.86953589e-02  -3.56495827e-01  -5.14335275e-01
#     1.98457405e-01  -2.56361932e-01]
#  [ -6.98417068e-01  -1.91254735e-01   2.43045017e-01   1.62485635e+00
#     1.34399071e-01   4.48576242e-01   9.65317804e-03   2.63324082e-01
#     1.71964037e+00   1.81242809e-01]
#  [ -3.06579232e-01   6.82186067e-01  -1.34259194e-01   8.21067274e-01
#     2.75264233e-01   2.36507237e-01  -5.18465579e-01   2.66540438e-01
#    -4.08049405e-01   4.00008321e-01]
#  [  6.39303386e-01  -3.73077720e-01  -3.93441260e-01   1.44391823e+00
#    -4.43163812e-01  -2.28796184e-01  -9.21004534e-01  -6.29425883e-01
#    -1.04850844e-01   5.76163709e-01]]
# ------------------------------------------------------------
# [[-1.36460924 -0.87841016]
#  [ 1.09924674 -1.08868659]]
# ------------------------------------------------------------
# [[  2.20000000e+01   2.20000000e+01   1.87626636e+00   1.27253577e-01
#     1.77748889e-01  -2.79777735e-01  -1.29324961e+00   1.18464601e+00
#     5.91014000e-03   3.91639054e-01]
#  [  2.20000000e+01   2.20000000e+01  -1.12289786e+00  -6.78496242e-01
#     4.25694019e-01   1.06920172e-02   8.05350840e-01  -7.63205171e-01
#     3.23375136e-01   4.49576497e-01]
#  [ -1.05841234e-01  -2.85442650e-01   1.01552248e+00  -1.69130695e+00
#    -1.15725782e-03   1.47837210e+00   6.69513494e-02   4.84152675e-01
#    -3.82478178e-01  -1.20722222e+00]
#  [  3.09620090e-02  -7.98057988e-02   5.60789108e-01  -1.78828037e+00
#     1.95968688e-01  -2.27404922e-01  -1.72476351e+00  -1.23355138e+00
#    -9.63081241e-01  -1.60457373e+00]
#  [  3.23699638e-02  -5.90677261e-01  -1.92470342e-01  -4.10615206e-01
#     3.35628651e-02  -1.41288012e-01  -3.40104014e-01   1.13289797e+00
#     6.52421772e-01   9.17639136e-01]
#  [  5.11528432e-01   1.64260656e-01  -7.84879684e-01  -2.26418376e-01
#    -8.90385628e-01  -3.38358462e-01   1.96581781e+00  -7.17314780e-02
#     1.23814851e-01   7.64766455e-01]
#  [  2.22003132e-01  -1.70102561e+00  -1.60155982e-01   1.10927892e+00
#    -1.14349496e+00  -6.86953589e-02  -3.56495827e-01  -5.14335275e-01
#     1.98457405e-01  -2.56361932e-01]
#  [ -6.98417068e-01  -1.91254735e-01   2.43045017e-01   1.62485635e+00
#     1.34399071e-01   4.48576242e-01   9.65317804e-03   2.63324082e-01
#     1.71964037e+00   1.81242809e-01]
#  [ -3.06579232e-01   6.82186067e-01  -1.34259194e-01   8.21067274e-01
#     2.75264233e-01   2.36507237e-01  -5.18465579e-01   2.66540438e-01
#    -4.08049405e-01   4.00008321e-01]
#  [  6.39303386e-01  -3.73077720e-01  -3.93441260e-01   1.44391823e+00
#    -4.43163812e-01  -2.28796184e-01  -9.21004534e-01  -6.29425883e-01
#    -1.04850844e-01   5.76163709e-01]]
# ------------------------------------------------------------
# [[  2.20000000e+01   2.20000000e+01   1.87626636e+00   1.27253577e-01
#     1.77748889e-01  -2.79777735e-01  -1.29324961e+00   1.18464601e+00
#     5.91014000e-03   3.91639054e-01]
#  [  2.20000000e+01   2.20000000e+01  -1.12289786e+00  -6.78496242e-01
#     4.25694019e-01   1.06920172e-02   8.05350840e-01  -7.63205171e-01
#     3.23375136e-01   4.49576497e-01]
#  [ -1.05841234e-01  -2.85442650e-01   1.01552248e+00  -1.69130695e+00
#    -1.15725782e-03   1.47837210e+00   6.69513494e-02   4.84152675e-01
#    -3.82478178e-01  -1.20722222e+00]
#  [  3.09620090e-02  -7.98057988e-02   5.60789108e-01  -1.78828037e+00
#     1.95968688e-01  -2.27404922e-01  -1.72476351e+00  -1.23355138e+00
#    -9.63081241e-01  -1.60457373e+00]
#  [  3.23699638e-02  -5.90677261e-01  -1.92470342e-01  -4.10615206e-01
#     3.35628651e-02  -1.41288012e-01  -3.40104014e-01   1.13289797e+00
#     6.52421772e-01   9.17639136e-01]
#  [  5.11528432e-01   1.64260656e-01  -7.84879684e-01  -2.26418376e-01
#    -8.90385628e-01  -3.38358462e-01   1.96581781e+00  -7.17314780e-02
#     1.23814851e-01   7.64766455e-01]
#  [  2.22003132e-01  -1.70102561e+00  -1.60155982e-01   1.10927892e+00
#    -1.14349496e+00  -6.86953589e-02  -3.56495827e-01  -5.14335275e-01
#     1.98457405e-01  -2.56361932e-01]
#  [ -6.98417068e-01  -1.91254735e-01   2.43045017e-01   1.62485635e+00
#     1.34399071e-01   4.48576242e-01   9.65317804e-03   2.63324082e-01
#     1.71964037e+00   1.81242809e-01]
#  [ -3.06579232e-01   6.82186067e-01  -1.34259194e-01   8.21067274e-01
#     2.75264233e-01   2.36507237e-01  -5.18465579e-01   2.66540438e-01
#    -4.08049405e-01   4.00008321e-01]
#  [  6.39303386e-01  -3.73077720e-01  -3.93441260e-01   1.44391823e+00
#    -4.43163812e-01  -2.28796184e-01  -9.21004534e-01  -6.29425883e-01
#    -1.04850844e-01   5.76163709e-01]]
# [[  2.20000000e+01   2.20000000e+01   1.87626636e+00   1.27253577e-01
#     1.77748889e-01  -2.79777735e-01  -1.29324961e+00   1.18464601e+00
#     5.91014000e-03   3.91639054e-01]
#  [  2.20000000e+01   2.20000000e+01  -1.12289786e+00  -6.78496242e-01
#     4.25694019e-01   1.06920172e-02   8.05350840e-01  -7.63205171e-01
#     3.23375136e-01   4.49576497e-01]
#  [ -1.05841234e-01  -2.85442650e-01   1.01552248e+00  -1.69130695e+00
#    -1.15725782e-03   1.47837210e+00   6.69513494e-02   4.84152675e-01
#    -3.82478178e-01  -1.20722222e+00]
#  [  3.09620090e-02  -7.98057988e-02   5.60789108e-01  -1.78828037e+00
#     1.95968688e-01  -2.27404922e-01  -1.72476351e+00  -1.23355138e+00
#    -9.63081241e-01  -1.60457373e+00]
#  [  3.23699638e-02  -5.90677261e-01  -1.92470342e-01  -4.10615206e-01
#     3.35628651e-02  -1.41288012e-01  -3.40104014e-01   1.13289797e+00
#     6.52421772e-01   9.17639136e-01]
#  [  5.11528432e-01   1.64260656e-01  -7.84879684e-01  -2.26418376e-01
#    -8.90385628e-01  -3.38358462e-01   1.96581781e+00  -7.17314780e-02
#     1.23814851e-01   7.64766455e-01]
#  [  2.22003132e-01  -1.70102561e+00  -1.60155982e-01   1.10927892e+00
#    -1.14349496e+00  -6.86953589e-02  -3.56495827e-01  -5.14335275e-01
#     1.98457405e-01  -2.56361932e-01]
#  [ -6.98417068e-01  -1.91254735e-01   2.43045017e-01   1.62485635e+00
#     1.34399071e-01   4.48576242e-01   9.65317804e-03   2.63324082e-01
#     1.71964037e+00   1.81242809e-01]
#  [ -3.06579232e-01   6.82186067e-01  -1.34259194e-01   8.21067274e-01
#     2.75264233e-01   2.36507237e-01  -5.18465579e-01   2.66540438e-01
#    -4.08049405e-01   4.00008321e-01]
#  [  6.39303386e-01  -3.73077720e-01  -3.93441260e-01   1.44391823e+00
#    -4.43163812e-01  -2.28796184e-01  -9.21004534e-01  -6.29425883e-01
#    -1.04850844e-01   5.76163709e-01]]
# ------------------------------------------------------------
# <dtype: 'float32_ref'>
# [[-1.26520741  0.15191878 -0.20875564 -1.3333143  -1.2195878   1.98078632
#   -0.37404546 -0.08800519  1.59779143  0.43416482]
#  [-0.49763143  1.67703199 -0.75461388  1.39778709 -0.59904957  1.07046878
#   -0.66246015  1.20341682 -0.91598368  0.02659398]
#  [ 0.15760519  1.58692122 -1.64905691 -0.82413578 -1.35249043 -0.05494672
#    0.1835352  -1.16055572  0.06633712 -0.60654968]
#  [ 0.44664332  1.32015193  1.19064319  1.04832637 -1.35440087  1.46269107
#   -0.73073471  1.79925811 -0.81201482 -0.52865958]
#  [ 0.86693144  0.58485049  0.74763191  0.16709246  0.87199962 -0.01152798
#   -0.45584357  1.63553476 -0.97385317  0.22666731]
#  [ 0.88356006  0.02941392  0.48637423  0.6196385   0.72364163  0.73068076
#    0.24449067 -1.82126474  0.43334225 -0.57979202]
#  [ 0.90533078 -0.46376067  1.15587556 -0.83500886  0.61234719  0.32427955
#    1.49570584 -0.48938131 -1.18225038  0.19612415]
#  [ 0.66482532  1.62603843  0.83715767  0.93517244 -1.04942799  1.2456522
#    1.45724511  0.82795757 -0.39859715 -1.37503803]
#  [ 0.84474677 -0.97584003 -0.23359941  0.02022619  0.15871523  0.07553232
#   -0.58795238 -1.66832328 -1.08443391 -0.25014412]
#  [ 0.00373921  0.82514638 -0.06235779  0.58550674 -0.90887821 -0.91024882
#    0.57362115 -1.92959452  1.00629067 -0.69716936]]
# None
# (10, 10)
# ------------------------------------------------------------
# [[ 1.  1.]
#  [ 2.  2.]]