TensorFlow2-创建Tensor(5)

偏执的太偏执、 2022-11-29 03:04 92阅读 0赞

### 文章目录 ###

*  创建Tensor
 *   *  From Numpy,List
     *   *  numpy创建数据并转为Tensor
         *  list 创建数据并转为Tensor
     *  tf.zeros
     *  tf.zeros\_like
     *  tf.ones
     *  tf.fill(自定义填充)
     *  tf.random.normal(服从指定正态分布的序列)
     *  tf.random.uniform(均匀分布)
     *  Random Permutation(随机排列)
     *  tf.random.constant
     *  Typical Dim Data(典型尺寸数据)
     *   *  Scalar 标量
     *  LOSS
     *  Vector
     *  Matrix
     *  Dim=3 Tensor
     *   *  pad\_sequences 序列预处理
     *  Dim=4 Tensor
     *  Dim=5 Tensor

# 创建Tensor #

*  from numpy, list
 *  zeros, ones
 *  fill
 *  random
 *  constant
 *  Application

## From Numpy,List ##

import tensorflow as tf
    import numpy as np

### numpy创建数据并转为Tensor ###

tf.convert_to_tensor(np.ones([2,3]))

<tf.Tensor: shape=(2, 3), dtype=float64, numpy=
    array([[1., 1., 1.],
           [1., 1., 1.]])>

tf.convert_to_tensor(np.zeros([2,3]))

<tf.Tensor: shape=(2, 3), dtype=float64, numpy=
    array([[0., 0., 0.],
           [0., 0., 0.]])>

### list 创建数据并转为Tensor ###

tf.convert_to_tensor([1,2])

<tf.Tensor: shape=(2,), dtype=int32, numpy=array([1, 2], dtype=int32)>

tf.convert_to_tensor([1,2.])

<tf.Tensor: shape=(2,), dtype=float32, numpy=array([1., 2.], dtype=float32)>

tf.convert_to_tensor([[1], [2.]])

<tf.Tensor: shape=(2, 1), dtype=float32, numpy=
    array([[1.],
           [2.]], dtype=float32)>

## tf.zeros ##

# 标量
    tf.zeros([])

<tf.Tensor: shape=(), dtype=float32, numpy=0.0>

# 1维
    tf.zeros([1])

<tf.Tensor: shape=(1,), dtype=float32, numpy=array([0.], dtype=float32)>

# 2维
    tf.zeros([2,2])

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[0., 0.],
           [0., 0.]], dtype=float32)>

# 3维
    tf.zeros([2,3,3])

<tf.Tensor: shape=(2, 3, 3), dtype=float32, numpy=
    array([[[0., 0., 0.],
            [0., 0., 0.],
            [0., 0., 0.]],
    
           [[0., 0., 0.],
            [0., 0., 0.],
            [0., 0., 0.]]], dtype=float32)>

## tf.zeros\_like ##

a = tf.zeros([2,3,3])
    a

# 根据传递进来的对象创建相同的shape数组
    tf.zeros_like(a)

a.shape

TensorShape([2, 3, 3])

# 与tf.zeros_like等同
    tf.zeros(a.shape)

## tf.ones ##

# 1维
    tf.ones(1)

<tf.Tensor: shape=(1,), dtype=float32, numpy=array([1.], dtype=float32)>

# 标量
    tf.ones([])

<tf.Tensor: shape=(), dtype=float32, numpy=1.0>

tf.ones([2])

<tf.Tensor: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>

tf.ones([2,3])

<tf.Tensor: shape=(2, 3), dtype=float32, numpy=
    array([[1., 1., 1.],
           [1., 1., 1.]], dtype=float32)>

tf.ones_like(a)

<tf.Tensor: shape=(2, 3, 3), dtype=float32, numpy=
    array([[[1., 1., 1.],
            [1., 1., 1.],
            [1., 1., 1.]],
    
           [[1., 1., 1.],
            [1., 1., 1.],
            [1., 1., 1.]]], dtype=float32)>

# 等同tf.ones_like
    tf.ones(a.shape)

## tf.fill(自定义填充) ##

tf.fill([2,2],value=0)

<tf.Tensor: shape=(2, 2), dtype=int32, numpy=
    array([[0, 0],
           [0, 0]], dtype=int32)>

tf.fill([2,2],value=0.)

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[0., 0.],
           [0., 0.]], dtype=float32)>

tf.fill([2,2],value=1)

<tf.Tensor: shape=(2, 2), dtype=int32, numpy=
    array([[1, 1],
           [1, 1]], dtype=int32)>

tf.fill([2,2],value=9)

<tf.Tensor: shape=(2, 2), dtype=int32, numpy=
    array([[9, 9],
           [9, 9]], dtype=int32)>

## tf.random.normal(服从指定正态分布的序列) ##

# 用于从“服从指定正态分布的序列”中随机取出指定个数的值
    # mean: 正态分布的平均值
    # stddev: 正态分布的标准差
    # mean: 正态分布的平均值
    # seed: 用于为分发创建随机种子
    tf.random.normal([2,2],mean=1,stddev=1)

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[ 1.2933502 , -0.03046083],
           [-0.20453262,  1.8908376 ]], dtype=float32)>

tf.random.normal([2,2])

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[ 0.3923446, -1.0170017],
           [-2.2753727,  1.1124496]], dtype=float32)>

# 截断正态分布
    # 从截断的正态分布中输出随机值，虽然同样是输出正态分布，但是它生成的值是在距离均值两个标准差范围之内的
    tf.random.truncated_normal([2,2],mean=0,stddev=1)

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[-0.09197889,  0.7197461 ],
           [-1.2158926 ,  0.64242524]], dtype=float32)>

## tf.random.uniform(均匀分布) ##

# 0-1之间均匀采样
    tf.random.uniform([2,2],minval=0,maxval=1)

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[0.59650195, 0.8330668 ],
           [0.3798474 , 0.04617167]], dtype=float32)>

# 0-100之间均匀采样
    tf.random.uniform([3,4],minval=0,maxval=100)

<tf.Tensor: shape=(3, 4), dtype=float32, numpy=
    array([[39.938354 , 20.946777 , 88.78066  , 65.06464  ],
           [67.6425   , 53.750908 , 63.23085  ,  1.2730479],
           [53.01423  , 93.832565 , 41.489197 , 97.90436  ]], dtype=float32)>

## Random Permutation(随机排列) ##

# 生成数据
    idx=tf.range(10)
    idx

<tf.Tensor: shape=(10,), dtype=int32, numpy=array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=int32)>

# 打散数据
    idx=tf.random.shuffle(idx)
    idx

<tf.Tensor: shape=(10,), dtype=int32, numpy=array([8, 0, 5, 2, 4, 1, 9, 3, 7, 6], dtype=int32)>

# 正太分布数据
    a=tf.random.normal([10,784])
    # 均匀分布数据
    b=tf.random.uniform([10], minval=0,maxval=10,dtype=tf.int32)

# 从params的axis维根据indices的参数值获取切片
    a=tf.gather(a,indices=idx)
    b=tf.gather(b,indices=idx)

a,b

(<tf.Tensor: shape=(10, 784), dtype=float32, numpy=
     array([[ 1.2042824 , -1.3816946 , -0.34105217, ...,  1.7581135 ,
             -0.0409344 , -0.7366836 ],
            [ 0.23826773,  0.7524342 ,  1.1266931 , ..., -0.23611061,
              0.03666063,  0.92886865],
            [-0.6004217 , -0.10280953,  0.3276803 , ..., -0.5391498 ,
             -0.08050299,  0.42966053],
            ...,
            [-0.9661765 ,  0.38981438, -1.3240955 , ...,  0.27162325,
              0.7889554 , -0.28455386],
            [-0.8107048 ,  0.18297605,  1.7718701 , ...,  0.1255566 ,
              0.8605448 , -0.01122306],
            [ 1.3737792 ,  0.00462327, -0.12949888, ...,  1.4805819 ,
             -0.79014236,  0.8239969 ]], dtype=float32)>,
     <tf.Tensor: shape=(10,), dtype=int32, numpy=array([0, 2, 6, 5, 7, 6, 8, 0, 2, 1], dtype=int32)>)

## tf.random.constant ##

# 从类似张量的对象创建常量张量
    tf.constant(1)

<tf.Tensor: shape=(), dtype=int32, numpy=1>

tf.constant(1.)

<tf.Tensor: shape=(), dtype=float32, numpy=1.0>

tf.constant([1])

<tf.Tensor: shape=(1,), dtype=int32, numpy=array([1], dtype=int32)>

tf.constant([1,2.])

<tf.Tensor: shape=(2,), dtype=float32, numpy=array([1., 2.], dtype=float32)>

# tf.constant([[1,2.],[3.]])
    tf.constant([[1,2.],[3.,2]])

<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
    array([[1., 2.],
           [3., 2.]], dtype=float32)>

## Typical Dim Data(典型尺寸数据) ##

*  \[\]
    
     *  标量 Scalar
     *  1,2,2.3
 *  \[d\]
 *  \[h,w\]
 *  \[b.len,vec\]
 *  \[b,h,w,c\]
 *  \[t,b,h,w,c\]
 *  …

### Scalar 标量 ###

*  \[\]
    
     *  1,2,2.3
 *  loss=mse(out,y)
 *  accuracy

## LOSS ##

# 生成均匀分布数据
    out=tf.random.uniform([4,10])
    out

<tf.Tensor: shape=(4, 10), dtype=float32, numpy=
    array([[0.7957759 , 0.6470387 , 0.85942817, 0.9195243 , 0.80446124,
            0.9332032 , 0.10627711, 0.0725143 , 0.88076556, 0.8592241 ],
           [0.92300546, 0.9134604 , 0.79779387, 0.44595373, 0.3783729 ,
            0.5998908 , 0.33701885, 0.00334263, 0.1695168 , 0.24933684],
           [0.91918683, 0.2754066 , 0.7670957 , 0.5798063 , 0.11099887,
            0.94225526, 0.90335643, 0.06025386, 0.1178335 , 0.60193515],
           [0.17284477, 0.09774017, 0.3873669 , 0.99980617, 0.3310578 ,
            0.60174   , 0.7005192 , 0.09624684, 0.8688313 , 0.3498112 ]],
          dtype=float32)>

# 生成标签数据
    y=tf.range(4)
    y

<tf.Tensor: shape=(4,), dtype=int32, numpy=array([0, 1, 2, 3], dtype=int32)>

# 热编码
    y=tf.one_hot(y,depth=10)
    y

<tf.Tensor: shape=(4, 10), dtype=float32, numpy=
    array([[1., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
           [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.],
           [0., 0., 1., 0., 0., 0., 0., 0., 0., 0.],
           [0., 0., 0., 1., 0., 0., 0., 0., 0., 0.]], dtype=float32)>

l o s s = ∑ ( y − o u t ) 2 loss=\\sum\_\{\}^\{\}(y-out)^2 loss=∑(y−out)2

# 计算标签和预测之间的均方误差
    loss=tf.keras.losses.mse(y,out)
    loss

<tf.Tensor: shape=(4,), dtype=float32, numpy=array([0.5093101 , 0.24023107, 0.34072328, 0.20383982], dtype=float32)>

# 计算张量维度上元素的平均值。
    loss=tf.reduce_mean(loss)
    loss

<tf.Tensor: shape=(), dtype=float32, numpy=0.32352605>

## Vector ##

*  Bias
    
     *  \[out\_dim\]
     *  y=X@W + b(bias)

from tensorflow.keras import layers, optimizers, datasets

# dense：全连接层
    # 相当于添加一个层，即初学的add_layer()函数
    # 8 -> 10
    net=layers.Dense(10)
    net.build((4,8))
    net.kernel

<tf.Variable 'kernel:0' shape=(8, 10) dtype=float32, numpy=
    array([[ 0.01621449, -0.14240894, -0.39780322,  0.12109792,  0.5365114 ,
             0.48510587,  0.10659033, -0.10367435, -0.16618788, -0.28067237],
           [ 0.4738747 ,  0.23055995, -0.28443712, -0.45606828, -0.2740961 ,
            -0.22440037, -0.179019  , -0.09799156, -0.5101716 , -0.12497845],
           [ 0.13550311, -0.30709794, -0.43564707,  0.5684767 ,  0.02459109,
             0.40579963,  0.41654438, -0.12095284,  0.5667572 ,  0.3747304 ],
           [ 0.11231458,  0.48574615,  0.24569553,  0.49175394, -0.23326564,
            -0.18449727, -0.2591746 ,  0.38994485,  0.47833967,  0.42860043],
           [-0.01137727,  0.39716053,  0.11606997,  0.57487595, -0.09682512,
            -0.25918382,  0.04865551, -0.02905029, -0.54025316, -0.39565778],
           [-0.18153116, -0.32627657,  0.53890324, -0.14481756,  0.5542736 ,
             0.2505871 , -0.15414259,  0.43023145, -0.01220387, -0.15518737],
           [ 0.04331303,  0.4358071 , -0.18711427, -0.24738216, -0.02527547,
             0.47772157,  0.35470355,  0.23595202,  0.28886396, -0.16550332],
           [-0.53486687, -0.16493705, -0.44315812,  0.11054051,  0.38065422,
            -0.14305219,  0.5382327 , -0.0840193 ,  0.46640956, -0.10600576]],
          dtype=float32)>

# b
    net.bias

<tf.Variable 'bias:0' shape=(10,) dtype=float32, numpy=array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)>

## Matrix ##

*  input x: \[b, vec\_dim\]
 *  weight: \[input\_dim, output\_dim\]

$  
W=\\begin\{bmatrix\}  
w1,1w1,2…w1R \\  
w2,1w2,2…w2R \\  
… \\  
ws,1ws,2…wsR  
\\end\{bmatrix\}  
$

x=tf.random.normal([4,784])

x.shape

TensorShape([4, 784])

# 784 -> 10
    net=layers.Dense(10)
    net.build((4,784))

net(x).shape

TensorShape([4, 10])

net.kernel.shape

TensorShape([784, 10])

net.bias.shape

TensorShape([10])

## Dim=3 Tensor ##

*  x: \[b, seq\_len, word\_dim\]
 *  \[b, 5, 5\]

# 加载数据集
    (X_train,y_train),(X_test,y_test)=tf.keras.datasets.imdb.load_data(num_words=10000)

# 样本长度
    print(X_train[0])
    print(len(X_train[0]))
    print(type(X_train[0]))

[1, 14, 22, 16, 43, 530, 973, 1622, 1385, 65, 458, 4468, 66, 3941, 4, 173, 36, 256, 5, 25, 100, 43, 838, 112, 50, 670, 2, 9, 35, 480, 284, 5, 150, 4, 172, 112, 167, 2, 336, 385, 39, 4, 172, 4536, 1111, 17, 546, 38, 13, 447, 4, 192, 50, 16, 6, 147, 2025, 19, 14, 22, 4, 1920, 4613, 469, 4, 22, 71, 87, 12, 16, 43, 530, 38, 76, 15, 13, 1247, 4, 22, 17, 515, 17, 12, 16, 626, 18, 2, 5, 62, 386, 12, 8, 316, 8, 106, 5, 4, 2223, 5244, 16, 480, 66, 3785, 33, 4, 130, 12, 16, 38, 619, 5, 25, 124, 51, 36, 135, 48, 25, 1415, 33, 6, 22, 12, 215, 28, 77, 52, 5, 14, 407, 16, 82, 2, 8, 4, 107, 117, 5952, 15, 256, 4, 2, 7, 3766, 5, 723, 36, 71, 43, 530, 476, 26, 400, 317, 46, 7, 4, 2, 1029, 13, 104, 88, 4, 381, 15, 297, 98, 32, 2071, 56, 26, 141, 6, 194, 7486, 18, 4, 226, 22, 21, 134, 476, 26, 480, 5, 144, 30, 5535, 18, 51, 36, 28, 224, 92, 25, 104, 4, 226, 65, 16, 38, 1334, 88, 12, 16, 283, 5, 16, 4472, 113, 103, 32, 15, 16, 5345, 19, 178, 32]
    218
    <class 'list'>

X_train.shape

(25000,)

### pad\_sequences 序列预处理 ###

*  sequences：浮点数或整数构成的两层嵌套列表
 *  maxlen：None或整数，为序列的最大长度。大于此长度的序列将被截短，小于此长度的序列将在后部填0.在命名实体识别任务中，主要是指句子的最大长度
 *  dtype：返回的numpy array的数据类型
 *  padding：‘pre’或‘post’，确定当需要补0时，在序列的起始还是结尾补
 *  truncating：‘pre’或‘post’，确定当需要截断序列时，从起始还是结尾截断
 *  value：浮点数，此值将在填充时代替默认的填充值0

# 序列预处理
    x_train=tf.keras.preprocessing.sequence.pad_sequences(X_train,maxlen=80)

print(x_train[0])
    print(len(x_train[0]))
    print(type(x_train[0]))

[  15  256    4    2    7 3766    5  723   36   71   43  530  476   26
      400  317   46    7    4    2 1029   13  104   88    4  381   15  297
       98   32 2071   56   26  141    6  194 7486   18    4  226   22   21
      134  476   26  480    5  144   30 5535   18   51   36   28  224   92
       25  104    4  226   65   16   38 1334   88   12   16  283    5   16
     4472  113  103   32   15   16 5345   19  178   32]
    80
    <class 'numpy.ndarray'>

x_train.shape

(25000, 80)

# 编码
    emb=embedding(x-train)

emb.shape

out=tf.rnn(emb[:4])

out.shape

## Dim=4 Tensor ##

*  Image: \[b, h, w, 3\]
    
     *  \[b,28,28,1\]
 *  feature maps: \[b, h, w, c\]
    
     *  \[b, h, w, 3\]

x=tf.random.normal((4,32,32,3))

# 卷积层
    net=layers.Conv2D(16,kernel_size=3)

net(x)

<tf.Tensor: shape=(4, 30, 30, 16), dtype=float32, numpy=
    array([[[[ 5.11408783e-02, -2.11419791e-01,  4.78755325e-01, ...,
              -7.09839880e-01, -1.62664950e-01, -3.69613528e-01],
             [ 1.04834512e-01,  4.45832849e-01,  3.67929399e-01, ...,
              -5.73501050e-01, -9.47547019e-01, -2.73244321e-01],
             [ 6.64293393e-02, -3.11614126e-01,  8.88543963e-01, ...,
              -5.91010869e-01,  8.83403420e-02, -6.25845611e-01],
             ...,
             [ 7.06125498e-01, -3.92723858e-01,  8.19511190e-02, ...,
               3.88073295e-01, -3.65344107e-01, -4.27012980e-01],
             [ 1.42910075e+00, -3.73372495e-01,  1.46203363e+00, ...,
              -2.29273185e-01,  1.61577657e-01, -3.62703115e-01],
             [ 1.52406737e-01,  9.83081281e-01, -9.29397166e-01, ...,
              -4.01978552e-01, -1.65519774e+00, -6.72560513e-01]],
    
            [[ 3.38849723e-01,  5.57779014e-01, -4.73486066e-01, ...,
               6.66084468e-01, -2.25408077e-01,  2.01416627e-01],
             [ 9.98801887e-02,  3.51594418e-01, -1.17291820e+00, ...,
              -4.78315353e-01,  1.71240568e-01,  2.29796067e-01],
             [ 1.28986561e+00,  4.30635571e-01,  2.68138438e-01, ...,
               3.85428101e-01, -1.30183734e-02,  7.93949187e-01],
             ...,
             [ 8.19463372e-01,  4.61028397e-01,  2.90335417e-02, ...,
              -2.65352666e-01, -9.04486954e-01,  4.05555129e-01],
             [ 2.95304179e-01,  7.62058318e-01, -6.67063117e-01, ...,
              -2.71042168e-01,  6.85480416e-01,  2.40263894e-01],
             [ 3.03411335e-01, -7.25541830e-01, -8.30628812e-01, ...,
              -7.50274420e-01, -1.50119960e-01, -3.95896226e-01]],
    
            [[-2.56495386e-01, -7.38008507e-03, -1.67473584e-01, ...,
               2.62954921e-01, -8.31158459e-01, -9.43797290e-01],
             [-3.24434787e-01,  7.00764358e-01, -7.61871561e-02, ...,
              -1.63662970e-01, -2.82550633e-01,  9.41351712e-01],
             [-4.85321730e-01, -2.02089727e-01, -5.27621686e-01, ...,
               2.42007345e-01,  2.84019470e-01,  2.84560677e-02],
             ...,
             [-2.54487991e-01,  3.96051168e-01, -6.56044304e-01, ...,
              -5.79203784e-01, -7.97876954e-01, -6.80033207e-01],
             [-4.36108887e-01, -4.78615195e-01,  1.12478447e+00, ...,
              -1.31961375e-01,  1.66062370e-01, -5.76823093e-02],
             [-1.49053439e-01,  1.66869834e-01, -5.55442452e-01, ...,
               3.28702837e-01, -1.06465340e+00,  2.32470438e-01]],
    
            ...,
    
            [[ 3.62408757e-01, -1.23813343e+00,  7.57653639e-02, ...,
               3.63634303e-02,  8.34123254e-01, -1.68675110e-01],
             [ 3.52181286e-01,  9.89539027e-01, -1.13494694e+00, ...,
               1.14149503e-01, -9.32057440e-01, -4.54324901e-01],
             [ 5.59438944e-01, -9.75797400e-02, -2.96774745e-01, ...,
              -9.09146965e-02, -1.40010104e-01, -2.78197140e-01],
             ...,
             [-6.19658291e-01, -1.24590620e-01, -4.91972193e-02, ...,
              -2.09599838e-01,  2.36650333e-01, -2.43948981e-01],
             [ 4.20804381e-01, -3.13818097e-01,  6.72074437e-01, ...,
               1.59845605e-01, -1.21049285e-01,  5.39864361e-01],
             [-1.12884209e-01, -9.41875696e-01,  3.67895700e-02, ...,
              -4.34767634e-01,  1.29640436e+00,  2.41676748e-01]],
    
            [[ 6.18064642e-01,  5.64845145e-01,  4.26679671e-01, ...,
               6.53841555e-01, -7.71579087e-01,  6.57868683e-02],
             [ 8.64152551e-01, -5.17681301e-01,  1.43626824e-01, ...,
              -1.70546189e-01,  8.96106884e-02,  4.54999506e-01],
             [-4.78937477e-01, -4.56122994e-01,  6.41848207e-01, ...,
              -1.19053459e+00,  1.20916940e-01,  2.71652102e-01],
             ...,
             [ 9.20986295e-01,  7.50426650e-01,  5.38934469e-01, ...,
               4.02694009e-02, -6.34055018e-01,  1.87944248e-02],
             [-1.02507330e-01,  1.09432077e+00,  9.18149129e-02, ...,
               2.04167053e-01, -1.80424526e-01, -2.74709523e-01],
             [-5.92706978e-01, -4.26084399e-01, -6.30943716e-01, ...,
              -2.03500003e-01,  5.43525159e-01, -7.80534931e-03]],
    
            [[ 3.69665354e-01,  5.48572429e-02,  7.47180283e-01, ...,
              -4.25332904e-01,  3.04142982e-01,  6.80612326e-02],
             [-2.93423682e-01,  2.24400461e-01, -1.75468147e-01, ...,
              -1.21893203e+00,  2.17650369e-01,  5.83284616e-01],
             [-4.92672741e-01,  7.17395246e-02, -1.09146468e-01, ...,
              -3.73993725e-01,  4.24882956e-02,  9.17238295e-02],
             ...,
             [ 6.62829280e-02, -1.58003271e-01, -1.24654186e+00, ...,
              -2.55260974e-01,  1.85372248e-01,  2.66077727e-01],
             [-1.95675135e-01,  5.16129769e-02, -3.96091282e-01, ...,
               5.76506853e-01, -4.20435786e-01, -3.86470616e-01],
             [ 3.98471832e-01,  2.92219281e-01,  4.01777089e-01, ...,
               1.58537793e+00, -9.73537207e-01,  2.93004245e-01]]],
    
    
           [[[-4.88410741e-01,  3.35286170e-01,  2.36457393e-01, ...,
               1.26244557e+00, -3.53360593e-01, -4.18140233e-01],
             [-3.03056449e-01,  7.47030601e-02,  5.35628617e-01, ...,
               3.22452217e-01, -1.48300707e-01,  2.03683227e-01],
             [ 8.70249152e-01,  6.30708635e-01, -1.95097223e-01, ...,
              -8.11083496e-01, -5.93419313e-01, -1.81713790e-01],
             ...,
             [-8.20982456e-02,  5.62634587e-01, -2.71823734e-01, ...,
               8.18851683e-03, -3.14377576e-01, -9.49833021e-02],
             [ 3.77315246e-02,  2.37164915e-01, -1.12752795e+00, ...,
               6.82983696e-01, -2.03109041e-01,  3.27775270e-01],
             [-3.42391729e-01,  4.12799299e-01, -1.03908813e+00, ...,
              -1.77690387e-01, -1.95123062e-01,  7.12375492e-02]],
    
            [[-6.04353905e-01, -4.86152679e-01,  7.74208426e-01, ...,
              -3.77850741e-01,  6.32216828e-03, -2.98906416e-01],
             [ 3.03069472e-01,  1.45399764e-01,  2.10653007e-01, ...,
               9.63516235e-01,  3.69627088e-01,  2.55155087e-01],
             [ 2.18334988e-01,  6.62136912e-01, -1.25356126e+00, ...,
              -2.10385859e-01,  6.05388105e-01,  5.24035275e-01],
             ...,
             [ 7.21290410e-01, -1.96054131e-01, -2.53071606e-01, ...,
               9.81350616e-03, -3.36887777e-01, -8.73216391e-01],
             [ 7.26956069e-01, -2.55486459e-01,  1.13994825e+00, ...,
              -7.81499624e-01, -9.61165130e-01, -3.59825432e-01],
             [ 1.11502385e+00, -7.02142417e-01,  1.41868222e+00, ...,
              -9.77690816e-01, -2.14777321e-01,  4.68507677e-01]],
    
            [[ 3.55722100e-01,  4.81817633e-01,  4.14835602e-01, ...,
               5.47380567e-01,  5.61807156e-01,  9.30055007e-02],
             [-5.89577481e-02,  3.67752552e-01, -2.10168734e-01, ...,
              -1.17150187e+00, -1.30239666e-01, -7.32884407e-01],
             [ 7.63650179e-01, -7.12301672e-01,  1.14955544e+00, ...,
              -1.16633463e+00,  5.95600665e-01,  9.20139626e-02],
             ...,
             [ 5.23808539e-01, -1.08179323e-01,  3.20617974e-01, ...,
              -7.01221347e-01,  3.13195109e-01,  6.03635669e-01],
             [-1.10597581e-01,  6.05741262e-01, -3.21878254e-01, ...,
              -9.76488471e-01,  8.60475838e-01, -3.45011771e-01],
             [ 1.34457603e-01,  6.90276265e-01, -4.44975734e-01, ...,
              -7.25347996e-01,  1.29672837e+00,  8.13677251e-01]],
    
            ...,
    
            [[-3.14656645e-01,  4.40141082e-01, -1.73647985e-01, ...,
              -9.47391808e-01, -7.14678824e-01, -1.45510659e-01],
             [ 2.90451586e-01,  4.85799819e-01,  3.68700057e-01, ...,
               3.72695416e-01,  9.44734871e-01,  1.65997118e-01],
             [ 9.50033963e-02, -7.15577960e-01, -1.43527782e+00, ...,
              -3.88930261e-01, -6.76119626e-01, -9.19058025e-01],
             ...,
             [ 4.16554570e-01,  8.66417825e-01, -4.78255659e-01, ...,
              -4.81267542e-01,  2.48430237e-01,  7.90232182e-01],
             [ 4.94117826e-01, -8.06771696e-01,  7.47416198e-01, ...,
              -7.30220854e-01,  1.18371248e+00, -5.31985700e-01],
             [ 7.23220631e-02, -2.15284348e-01, -1.92686126e-01, ...,
              -8.46448898e-01, -5.97360777e-03, -4.28196698e-01]],
    
            [[-2.25013599e-01, -3.48351777e-01, -5.54229319e-01, ...,
              -1.17885478e-01,  6.86471090e-02, -7.48188347e-02],
             [-6.46478087e-02, -7.97209620e-01,  8.16888452e-01, ...,
               1.15718633e-01,  1.01453686e+00, -2.39759922e-01],
             [-5.25327086e-01,  5.88182211e-01, -9.02729273e-01, ...,
               7.62883961e-01, -1.38593817e+00, -4.31858391e-01],
             ...,
             [ 7.87195325e-01, -2.07848758e-01, -1.93571016e-01, ...,
               7.28979111e-01,  5.80614686e-01,  3.75502616e-01],
             [-2.03104734e-01,  7.44872868e-01,  1.03454679e-01, ...,
              -1.42501622e-01,  2.46956795e-01,  1.22676164e-01],
             [ 7.62553990e-01, -5.06239831e-01,  1.93014458e-01, ...,
               5.26253104e-01,  2.82648057e-01,  8.88811767e-01]],
    
            [[-1.04494788e-01, -6.20330453e-01, -2.88904518e-01, ...,
              -9.90672529e-01,  2.64258742e-01,  2.14836478e-01],
             [ 3.59708428e-01, -9.44501907e-02,  9.55824256e-01, ...,
              -9.02668655e-01,  1.85327336e-01,  1.01691764e-03],
             [-3.08903068e-01,  1.56039998e-01,  6.49919152e-01, ...,
               3.52934897e-01, -4.29365396e-01, -1.05594814e+00],
             ...,
             [ 5.94020307e-01,  3.79775882e-01,  8.62827957e-01, ...,
              -8.00387859e-01, -5.81361890e-01,  8.48889947e-02],
             [-2.51786768e-01, -3.84436883e-02,  3.43284369e-01, ...,
               1.34449214e-01,  1.51265606e-01, -3.25558543e-01],
             [-5.40423274e-01,  8.11421871e-01, -1.04027975e+00, ...,
               1.75916687e-01, -3.55451852e-01, -4.18996513e-01]]],
    
    
           [[[ 4.03016716e-01, -9.42383185e-02,  3.89042109e-01, ...,
              -6.66733503e-01,  7.01597154e-01,  9.04736102e-01],
             [-8.01422894e-01,  9.61738050e-01, -1.64667815e-01, ...,
              -5.52261829e-01, -6.96025670e-01, -1.16403568e+00],
             [-3.79517712e-02,  1.41728476e-01, -3.10020149e-01, ...,
               1.44183421e+00,  5.36346138e-02, -1.90599233e-01],
             ...,
             [-5.14048576e-01, -2.21568689e-01,  5.31760991e-01, ...,
              -1.03457296e+00,  2.44072124e-01, -5.40369947e-04],
             [-5.53254724e-01,  1.03404298e-01, -8.44387431e-03, ...,
               5.00614405e-01,  2.78823525e-01,  2.91976810e-01],
             [-6.95665359e-01,  5.69382131e-01, -1.19335508e+00, ...,
              -2.95902044e-01,  3.53888348e-02,  5.01627335e-03]],
    
            [[ 1.20825207e+00, -3.29682589e-01, -3.54656160e-01, ...,
              -8.61604214e-02, -2.49787807e-01,  5.63491404e-01],
             [ 6.02723479e-01, -8.93569291e-02,  5.49378812e-01, ...,
               5.55969119e-01, -4.93603408e-01,  1.42602459e-01],
             [ 4.12537068e-01, -8.80664825e-01,  6.61745310e-01, ...,
              -7.31622279e-01, -7.21969008e-01, -1.03079927e+00],
             ...,
             [-6.26841545e-01, -3.40548962e-01,  3.20449062e-02, ...,
              -7.87570834e-01,  9.43054736e-01, -1.60972886e-02],
             [ 8.26191306e-01, -5.19545734e-01,  5.71801364e-01, ...,
               2.79021472e-01, -3.34796578e-01, -8.44013095e-01],
             [ 3.60502213e-01,  4.38319683e-01,  2.13991180e-01, ...,
              -6.36544347e-01, -5.06267473e-02, -5.51769555e-01]],
    
            [[-8.10957193e-01,  1.13378942e+00, -1.47388566e+00, ...,
               2.66789883e-01,  7.00945854e-02,  4.77081358e-01],
             [ 2.62345165e-01, -4.42802548e-01,  1.32407904e-01, ...,
              -1.38010776e+00, -1.99280515e-01,  8.50553453e-01],
             [ 6.35858774e-01,  1.65538931e+00, -2.15716064e-01, ...,
              -5.80295384e-01, -5.62521100e-01,  7.77037382e-01],
             ...,
             [ 5.82785085e-02, -4.98133540e-01, -3.92823726e-01, ...,
               5.99813223e-01,  5.84782600e-01,  1.82665259e-01],
             [ 1.19108522e+00,  4.52362090e-01, -2.45918483e-02, ...,
              -2.15084866e-01, -2.15071514e-01,  1.52354971e-01],
             [ 5.26105404e-01,  2.52852559e-01,  3.06775961e-02, ...,
              -1.40375543e+00,  1.26784432e+00,  4.97334570e-01]],
    
            ...,
    
            [[ 7.99124658e-01,  6.19785070e-01, -2.05564678e-01, ...,
               8.10738921e-01, -1.40358889e+00,  1.31820530e-01],
             [-4.84217465e-01, -4.65817660e-01, -1.86626345e-01, ...,
              -2.83500314e-01, -1.27768144e-01, -8.91454667e-02],
             [ 3.24485391e-01, -8.31413984e-01,  2.12071046e-01, ...,
              -3.96526724e-01, -7.63818026e-01, -4.76207286e-01],
             ...,
             [-2.97311485e-01, -1.00890353e-01, -4.08464611e-01, ...,
              -3.07795316e-01,  3.18488568e-01,  3.35744619e-01],
             [-4.40074831e-01,  4.96284395e-01, -5.66527307e-01, ...,
               1.37263402e-01, -6.25700802e-02, -2.86034942e-01],
             [-2.77784169e-01, -2.70140618e-01,  9.17928815e-02, ...,
               1.35965988e-01,  2.35500276e-01, -2.41373680e-04]],
    
            [[-3.21114600e-01,  3.98868248e-02,  3.31562501e-03, ...,
              -1.06464684e+00, -1.07360594e-02,  1.61534380e-02],
             [-5.94698310e-01, -7.75814295e-01, -8.55177194e-02, ...,
               6.45053446e-01, -5.99738657e-02,  3.18236798e-01],
             [ 5.18429339e-01,  6.83645844e-01,  5.31058192e-01, ...,
              -6.40040517e-01, -2.74345487e-01,  6.71724200e-01],
             ...,
             [-2.59246260e-01, -9.52060878e-01,  1.99519381e-01, ...,
              -7.06413329e-01,  5.07094800e-01,  1.99018568e-01],
             [-6.45661831e-01, -2.64860898e-01,  4.11892146e-01, ...,
              -2.21386373e-01, -1.03395559e-01, -4.46663290e-01],
             [ 7.03365728e-02, -1.71302333e-01,  6.37728274e-01, ...,
               5.99841118e-01, -3.00094217e-01, -1.68467209e-01]],
    
            [[-1.63223729e-01,  5.45847535e-01,  8.18708718e-01, ...,
              -7.14481771e-01,  1.95999846e-01, -2.45239045e-02],
             [-8.57500196e-01, -1.14799410e-01, -3.29935670e-01, ...,
               1.40195891e-01, -1.25240743e+00, -1.42649800e-01],
             [ 3.91972065e-01,  2.59413898e-01,  3.64314884e-01, ...,
               8.55417967e-01,  1.25591683e+00,  5.47372282e-01],
             ...,
             [ 9.93716195e-02,  7.77123153e-01,  8.93170595e-01, ...,
               3.03051144e-01,  7.42947683e-02,  7.33308613e-01],
             [ 3.11880819e-02, -2.95177728e-01,  2.06190944e-01, ...,
               1.53311670e+00,  4.61079597e-01,  3.99204530e-02],
             [-5.59754610e-01,  2.39114627e-01, -1.36300159e+00, ...,
              -4.93013799e-01, -7.12665096e-02,  7.62722492e-02]]],
    
    
           [[[-7.07550526e-01, -3.03240091e-01,  6.60755455e-01, ...,
              -7.90750161e-02, -3.65321159e-01, -4.30483460e-01],
             [-1.86911270e-01,  2.39894748e-01, -9.40757334e-01, ...,
               3.57000470e-01, -2.11175028e-02, -1.62767589e-01],
             [ 5.51772773e-01,  1.32477924e-01,  4.75931138e-01, ...,
               6.03891313e-01,  3.50777097e-02, -2.31402159e-01],
             ...,
             [ 3.60434175e-01, -2.40440086e-01,  1.36279285e-01, ...,
               8.35516527e-02,  8.36447299e-01, -5.07031977e-02],
             [-2.87243992e-01,  5.44507802e-01, -7.85944313e-02, ...,
              -3.60828161e-01,  5.22470474e-02,  2.05641940e-01],
             [-6.79957271e-01, -2.88661897e-01, -6.68929160e-01, ...,
              -8.08060616e-02,  3.17035764e-01,  3.03957164e-02]],
    
            [[-1.76532626e-01,  5.74718297e-01, -6.78593397e-01, ...,
               7.33274281e-01, -1.62762314e-01, -3.74133945e-01],
             [ 6.47938073e-01, -8.46699327e-02, -2.87556887e-01, ...,
               5.72368860e-01,  2.43657812e-01,  6.66255951e-01],
             [ 4.25566167e-01, -6.37027442e-01,  2.88296103e-01, ...,
              -5.30690730e-01,  5.80780327e-01,  2.29448169e-01],
             ...,
             [ 1.95712298e-01,  3.00102949e-01, -6.41289175e-01, ...,
               7.47099340e-01, -5.37887156e-01, -3.03677708e-01],
             [ 5.38146235e-02, -4.94787879e-02,  1.87186092e-01, ...,
               6.59565553e-02, -3.63303535e-02, -2.02135652e-01],
             [ 1.06148578e-01,  2.69699812e-01, -1.33452356e-01, ...,
               7.64168203e-01, -9.29973841e-01,  6.16707318e-02]],
    
            [[ 5.97640455e-01,  3.70381474e-01, -1.05833687e-01, ...,
               1.04272676e+00, -4.54665720e-01, -7.94478431e-02],
             [ 2.62423337e-01,  2.42638126e-01,  3.63827407e-01, ...,
              -1.93119079e-01,  6.85722753e-02,  3.13739687e-01],
             [-2.03977063e-01, -8.53810668e-01,  1.41490296e-01, ...,
              -7.12618008e-02,  2.41974235e-01,  1.75281800e-02],
             ...,
             [-3.56253654e-01, -2.64765304e-02, -1.70551449e-01, ...,
              -1.18611500e-01, -4.53533858e-01, -1.33774772e-01],
             [ 2.49976337e-01,  6.65123165e-02,  3.74623120e-01, ...,
               1.98802855e-02,  2.49585107e-01,  4.90735173e-01],
             [ 1.00049064e-01,  1.08251944e-01, -4.02322143e-01, ...,
               5.90216041e-01, -9.96080697e-01,  2.94348657e-01]],
    
            ...,
    
            [[-2.33743116e-01,  3.79105168e-03,  1.00059927e+00, ...,
              -3.50684762e-01, -2.03996941e-01,  5.43174446e-01],
             [-1.31496513e+00, -7.68835992e-02,  9.21740979e-02, ...,
               1.82122439e-01,  3.92907321e-01,  4.70603764e-01],
             [-1.91511422e-01, -3.12377542e-01, -2.02807784e-01, ...,
               4.74840961e-02,  3.90858687e-02,  5.20986259e-01],
             ...,
             [ 3.92100602e-01, -2.56313950e-01,  4.40316498e-01, ...,
              -7.95090020e-01,  9.58722889e-01,  6.68692514e-02],
             [-2.37561002e-01, -2.45921806e-01, -4.79896784e-01, ...,
               3.05862486e-01,  4.18518394e-01, -7.09717393e-01],
             [ 9.63735729e-02,  5.26934683e-01, -2.43123412e-01, ...,
              -2.29806781e-01,  8.54906738e-01,  1.45679355e-01]],
    
            [[-6.50098324e-01,  1.00708961e+00,  4.81430953e-03, ...,
              -4.03477103e-01,  5.49487919e-02, -5.28567910e-01],
             [-7.36448050e-01,  2.37548098e-01,  6.89756870e-01, ...,
               1.44244343e-01,  4.32573399e-03, -1.50717488e-02],
             [-3.26949120e-01,  1.15860653e+00, -4.18980062e-01, ...,
               6.63382530e-01,  2.57124096e-01,  5.86564839e-01],
             ...,
             [ 5.46750352e-02,  3.76966506e-01,  1.79619983e-01, ...,
               7.27700114e-01,  7.69920886e-01,  3.30334604e-01],
             [-3.29259604e-01, -7.28786111e-01,  4.86835659e-01, ...,
               3.71421605e-01,  6.50456697e-02, -1.75632238e-01],
             [-2.41953239e-01, -5.99552631e-01,  2.57653624e-01, ...,
               3.01961869e-01,  9.32633221e-01, -5.71823642e-02]],
    
            [[ 1.64552823e-01, -1.26616490e+00,  1.83328986e-01, ...,
               7.92825967e-02,  9.05444622e-01,  2.93801337e-01],
             [ 1.69471517e-01,  1.89377040e-01, -3.94473910e-01, ...,
               6.73406303e-01, -2.55624741e-01, -5.90663373e-01],
             [-2.28453115e-01,  1.53474778e-01,  7.06688702e-01, ...,
               1.02944827e+00, -9.44124758e-02, -4.60808694e-01],
             ...,
             [ 2.38406166e-01, -1.16364264e+00,  9.53676283e-01, ...,
               5.98508120e-01,  8.32367301e-01,  2.21683785e-01],
             [-2.71952063e-01, -6.16223276e-01,  4.03875411e-01, ...,
               2.75555700e-02,  1.19690061e-01,  1.28521711e-01],
             [ 1.00875773e-01, -5.86154759e-01,  5.06292939e-01, ...,
               1.41296101e+00,  1.24331325e-01, -2.65127104e-02]]]],
          dtype=float32)>

## Dim=5 Tensor ##

*  Single task: \[b, h, w, 3\]
 *  meta-learning:
 *  \[task\_b, b, h, w, 3\]