NumPy庫是用于科學(xué)計(jì)算的一個(gè)開源Python擴(kuò)充程序庫，是其他數(shù)據(jù)分析包的基礎(chǔ)包,，它為Python提供了高性能數(shù)組與矩陣運(yùn)算處理能力,。

2.1 ndarray多維數(shù)組

NumPy庫為Python帶來了真正的ndarray多維數(shù)組功能。ndarray對(duì)象是一個(gè)快速而靈活的數(shù)據(jù)集容器

2.1.1 創(chuàng)建ndarray數(shù)組

NumPy庫能將序列數(shù)據(jù)（列表,、元組,、數(shù)組或其他序列類型）轉(zhuǎn)換為ndarray數(shù)組

import numpy as np
data1 = [5,7,9,20] #列表
arr1 = np.array(data1)
arr1
array([ 5,  7,  9, 20])
data2 = (5,7,9,20) #元組 
arr2 = np.array(data2)
arr2
array([ 5,  7,  9, 20])
data3 = [[1,2,3,4],[5,6,7,8]] #對(duì)于多維數(shù)組的創(chuàng)建，使用嵌套序列數(shù)據(jù)即可完成
arr3 = np.array(data3)
arr3
array([[1, 2, 3, 4],
       [5, 6, 7, 8]])
arr3.shape #shape是ndarray維度大小的元組
(2, 4)
arr3.dtype #dtype是解釋說明ndarray數(shù)據(jù)類型的對(duì)象
dtype('int32')
data4 = [1,2,2,3.45,5] #當(dāng)序列中有整數(shù)和浮點(diǎn)數(shù)時(shí),，NumPy會(huì)把數(shù)組的dtype定義為浮點(diǎn)數(shù)據(jù)類型
arr4 = np.array(data4)
arr4
array([1.  , 2.  , 2.  , 3.45, 5.  ])
arr4.dtype
dtype('float64')

特殊數(shù)組

np.zeros(8) #zeros函數(shù)可以創(chuàng)建指定長度或形狀的全0數(shù)組
array([0., 0., 0., 0., 0., 0., 0., 0.])
np.ones(4) #ones函數(shù)可以創(chuàng)建指定長度或形狀的全1數(shù)組
array([1., 1., 1., 1.])
np.ones((4,6)) # ones函數(shù)可以創(chuàng)建指定長度或形狀的全1數(shù)組
array([[1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1.]])
np.empty((2,3,2))  #mpty函數(shù)可以創(chuàng)建一個(gè)沒有具體值的數(shù)組（即垃圾值）

array([[[9.42595773e-312, 2.81617418e-322],
        [0.00000000e+000, 0.00000000e+000],
        [6.23060065e-307, 2.42336543e-057]],

       [[7.11697381e-091, 2.03997512e+184],
        [1.53389691e-052, 4.74680389e+174],
        [6.48224660e+170, 4.93432906e+257]]])
np.arange(10) #arange函數(shù)類似于Python的內(nèi)置函數(shù)range,，但是arange函數(shù)主要用于創(chuàng)建數(shù)組
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
arr3
array([[1, 2, 3, 4],
       [5, 6, 7, 8]])
arr5 = np.ones_like(arr3) #ones_like函數(shù)可以根據(jù)傳入的數(shù)組形狀和dtype創(chuàng)建全1數(shù)組。
arr5
array([[1, 1, 1, 1],
       [1, 1, 1, 1]])
arr5.dtype
dtype('int32')

2.1.2 ndarray對(duì)象屬性

data = [[2,4,5],[3,5,7]]
arr = np.array(data)
arr
array([[2, 4, 5],
       [3, 5, 7]])
arr.ndim #數(shù)據(jù)軸的個(gè)數(shù)
2
arr.size #元組的總個(gè)數(shù)
6
arr.itemsize #每個(gè)元素的字節(jié)大小
#arr數(shù)組的數(shù)據(jù)類型是int32位的,，
#對(duì)于計(jì)算機(jī)而言,，1個(gè)字節(jié)是8位，所以arr的itemsize屬性值為4,。
4
arr.dtype #數(shù)據(jù)類型
dtype('int32')

2.1.3 ndarray數(shù)據(jù)類型

數(shù)組的數(shù)據(jù)類型有很多,，讀者只需要記住最常見的幾種數(shù)據(jù)類型即可，如浮點(diǎn)數(shù)（float）,、整數(shù)（int）,、復(fù)數(shù)（complex）、布爾值（bool）,、字符串（string_）和Python對(duì)象（object）,。

arr1 = np.arange(6)
arr1
array([0, 1, 2, 3, 4, 5])
arr1.dtype
dtype('int32')
arr2 = arr1.astype(np.float64)
arr2
array([0., 1., 2., 3., 4., 5.])
arr2.dtype
dtype('float64')
arr3 = arr1.astype("string_") #對(duì)于創(chuàng)建好的ndarray，可通過astype方法進(jìn)行數(shù)據(jù)類型的轉(zhuǎn)換
arr3
array([b'0', b'1', b'2', b'3', b'4', b'5'], dtype='|S11')
arr3.dtype

dtype('S11')
arr3
array([b'0', b'1', b'2', b'3', b'4', b'5'], dtype='|S11')
arr3.astype(np.int32) #如果數(shù)組是字符串類型且全是數(shù)字的話,，也可以通過astype方法將其轉(zhuǎn)換為數(shù)值類型
array([0, 1, 2, 3, 4, 5])
arr = np.array(['2','hello'])
arr
array(['2', 'hello'], dtype='<U5')
arr.astype("int32") #但如果字符串中有字符時(shí),，轉(zhuǎn)換時(shí)就會(huì)報(bào)錯(cuò)，如圖2.15所示,。
---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)

~\AppData\Local\Temp/ipykernel_30528/2154082202.py in <module>
----> 1 arr.astype("int32") #但如果字符串中有字符時(shí),，轉(zhuǎn)換時(shí)就會(huì)報(bào)錯(cuò)，如圖2.15所示,。


ValueError: invalid literal for int() with base 10: 'hello'
arr1 = np.arange(10)
arr1.dtype
dtype('int32')
arr2 = np.ones(5)
arr2.dtype
dtype('float64')
arr3= arr1.astype(arr2.dtype) #astype方法也可以通過另外一個(gè)數(shù)組的dtype進(jìn)行轉(zhuǎn)換
arr3.dtype
dtype('float64')
arr = np.arange(3)
arr.dtype
dtype('int32')
arr.astype("float64")  #astype方法會(huì)創(chuàng)建一個(gè)新的數(shù)組,，并不會(huì)改變?cè)袛?shù)組的數(shù)據(jù)類型
array([0., 1., 2.])
arr
array([0, 1, 2])

2.1.4 數(shù)組變換

1．?dāng)?shù)組重塑

arr = np.arange(9)
arr
array([0, 1, 2, 3, 4, 5, 6, 7, 8])
arr.reshape((3,3)) #對(duì)于定義好的數(shù)組，可以通過reshape方法改變其數(shù)據(jù)維度,。傳入的參數(shù)為新維度的元組,，
array([[0, 1, 2],
       [3, 4, 5],
       [6, 7, 8]])
arr = np.array([[3,4,5],[1,2,3]])
arr.reshape((3,2)) #多維數(shù)組也可以被重塑
array([[3, 4],
       [5, 1],
       [2, 3]])
arr = np.arange(12)
arr.reshape((3,-1)) #reshape的參數(shù)中的一維參數(shù)可以設(shè)置為-1,，表示數(shù)組的維度可以通過數(shù)據(jù)本身來推斷
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
arr = np.arange(10).reshape((5,2))
arr
array([[0, 1],
       [2, 3],
       [4, 5],
       [6, 7],
       [8, 9]])
arr.ravel() #與reshape相反的方法是數(shù)據(jù)散開（ravel）數(shù)據(jù)或扁平化（flatten）
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
arr.flatten()
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
arr #數(shù)據(jù)重塑不會(huì)改變?cè)瓟?shù)組,。
array([[0, 1],
       [2, 3],
       [4, 5],
       [6, 7],
       [8, 9]])

2．?dāng)?shù)組合并

arr1 = np.arange(12).reshape((3,-1))
arr1
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
arr2 = np.arange(12,24).reshape((3,4))
arr2
array([[12, 13, 14, 15],
       [16, 17, 18, 19],
       [20, 21, 22, 23]])
np.concatenate([arr1,arr2],axis = 0)
#數(shù)組合并用于幾個(gè)數(shù)組間的操作,，concatenate方法通過指定軸方向，將多個(gè)數(shù)組合并在一起
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15],
       [16, 17, 18, 19],
       [20, 21, 22, 23]])
np.concatenate([arr1,arr2],axis = 1)
array([[ 0,  1,  2,  3, 12, 13, 14, 15],
       [ 4,  5,  6,  7, 16, 17, 18, 19],
       [ 8,  9, 10, 11, 20, 21, 22, 23]])
np.vstack((arr1,arr2)) 
#NumPy中提供了幾個(gè)比較簡單易懂的方法,，也可以進(jìn)行數(shù)組合并,，如vstack和hstack
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15],
       [16, 17, 18, 19],
       [20, 21, 22, 23]])
np.hstack((arr1,arr2))
array([[ 0,  1,  2,  3, 12, 13, 14, 15],
       [ 4,  5,  6,  7, 16, 17, 18, 19],
       [ 8,  9, 10, 11, 20, 21, 22, 23]])

3．?dāng)?shù)組拆分

arr = np.arange(12).reshape((6,2))
arr
array([[ 0,  1],
       [ 2,  3],
       [ 4,  5],
       [ 6,  7],
       [ 8,  9],
       [10, 11]])
np.split(arr,[2,4])
#數(shù)組拆分是數(shù)組合并的相反操作，通過split方法可以將數(shù)組拆分為多個(gè)數(shù)組
[array([[0, 1],
        [2, 3]]),
 array([[4, 5],
        [6, 7]]),
 array([[ 8,  9],
        [10, 11]])]

4．?dāng)?shù)組轉(zhuǎn)置和軸對(duì)換

arr = np.arange(12).reshape(3,4)
arr
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
arr.transpose((1,0))
#轉(zhuǎn)置是數(shù)組重塑的一種特殊形式,，可以通過transpose方法進(jìn)行轉(zhuǎn)置
#transpose方法需要傳入軸編號(hào)組成的元組,，這樣就完成了數(shù)組的轉(zhuǎn)置
array([[ 0,  4,  8],
       [ 1,  5,  9],
       [ 2,  6, 10],
       [ 3,  7, 11]])
arr.T
#除了使用transpose方法外，數(shù)組有著T屬性,，可用于數(shù)組的轉(zhuǎn)置
array([[ 0,  4,  8],
       [ 1,  5,  9],
       [ 2,  6, 10],
       [ 3,  7, 11]])
arr = np.arange(16).reshape((2,2,4))
arr
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7]],

       [[ 8,  9, 10, 11],
        [12, 13, 14, 15]]])
arr.swapaxes(1,2) #ndarray的swapaxes方法用于軸對(duì)換
array([[[ 0,  4],
        [ 1,  5],
        [ 2,  6],
        [ 3,  7]],

       [[ 8, 12],
        [ 9, 13],
        [10, 14],
        [11, 15]]])

2.1.5 NumPy的隨機(jī)數(shù)函數(shù)

arr = np.random.randint(100,200,size=(5,4))
#可以通過randint函數(shù)生成整數(shù)隨機(jī)數(shù)
arr
array([[148, 112, 119, 144],
       [194, 123, 131, 172],
       [147, 104, 168, 165],
       [101, 116, 131, 195],
       [160, 149, 194, 173]])
arr = np.random.randn(2,3,5)
#如randn函數(shù),，例如，生成平均數(shù)為0,，標(biāo)準(zhǔn)差為1的正態(tài)分布的隨機(jī)數(shù)
arr
array([[[-0.48274113,  0.07529806, -1.81740745,  1.286357  ,
         -2.55607301],
        [ 1.07998934, -0.31404883, -2.297629  , -0.66385757,
          0.95792075],
        [-0.51302191, -2.53759601, -0.26012523,  0.71636383,
          2.57003766]],

       [[ 1.9540761 ,  1.07840586, -0.24478286,  0.8714337 ,
          0.70059484],
        [-0.14210736,  0.09356213, -2.96823017, -0.55159717,
         -1.63454057],
        [-0.31135985, -1.52214541,  1.18820464,  1.96057804,
         -0.59961429]]])
arr = np.random.normal(4,5,size=(3,5))
#通過normal函數(shù)生成指定均值和標(biāo)準(zhǔn)差的正態(tài)分布的數(shù)組 
arr
array([[ 1.03057369e+01,  2.59867561e-01,  1.43530708e+01,
         5.99577642e+00,  1.41228837e-01],
       [-1.56586338e+00,  1.99540963e+00,  7.90100440e+00,
         1.64716969e+00, -7.06954944e+00],
       [ 4.79296269e+00,  5.51927601e-03,  7.69985462e+00,
        -8.28181420e+00,  7.95558205e+00]])
arr = np.random.randint(100,200,size=(5,4))
arr
array([[107, 142, 175, 160],
       [122, 124, 178, 193],
       [149, 123, 198, 100],
       [170, 165, 165, 139],
       [116, 151, 111, 189]])
np.random.permutation(arr) #對(duì)一個(gè)序列隨機(jī)排序,，不改變?cè)瓟?shù)組
array([[170, 165, 165, 139],
       [116, 151, 111, 189],
       [149, 123, 198, 100],
       [122, 124, 178, 193],
       [107, 142, 175, 160]])
arr
array([[107, 142, 175, 160],
       [122, 124, 178, 193],
       [149, 123, 198, 100],
       [170, 165, 165, 139],
       [116, 151, 111, 189]])
np.random.shuffle(arr) #對(duì)一個(gè)序列隨機(jī)排序，改變?cè)瓟?shù)組
arr
array([[170, 165, 165, 139],
       [122, 124, 178, 193],
       [107, 142, 175, 160],
       [149, 123, 198, 100],
       [116, 151, 111, 189]])

2.2 數(shù)組的索引和切片

2.2.1 數(shù)組的索引

import numpy as np
arr = np.arange(10)
arr
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
arr[3] #一維數(shù)組的索引類似Python列表
3
arr[-1]
9
arr[2] = 123
arr
array([  0,   1, 123,   3,   4,   5,   6,   7,   8,   9])
arr = np.arange(15).reshape(3,5)
arr
array([[ 0,  1,  2,  3,  4],
       [ 5,  6,  7,  8,  9],
       [10, 11, 12, 13, 14]])
arr[0]  # 對(duì)于二維數(shù)組,，可在單個(gè)或多個(gè)軸向上完成切片
array([0, 1, 2, 3, 4])
arr[2]
array([10, 11, 12, 13, 14])
arr[0][3] #如果需要獲取各個(gè)元素
3

2.2.2 數(shù)組的切片

arr = np.arange(6)
arr

array([0, 1, 2, 3, 4, 5])
arr[2:5]
array([2, 3, 4])
arr = np.arange(12).reshape(4,3)
arr
array([[ 0,  1,  2],
       [ 3,  4,  5],
       [ 6,  7,  8],
       [ 9, 10, 11]])
arr[2:] #當(dāng)在中括號(hào)中輸入一個(gè)參數(shù)時(shí),，數(shù)組就會(huì)按照0軸（也就是第一軸）方向進(jìn)行切片
array([[ 6,  7,  8],
       [ 9, 10, 11]])
arr

array([[ 0,  1,  2],
       [ 3,  4,  5],
       [ 6,  7,  8],
       [ 9, 10, 11]])
arr[:,1]
# 通過傳入多個(gè)參數(shù)（可以是整數(shù)索引和切片），即可完成任意數(shù)據(jù)的獲取
# 只有使用冒號(hào)才會(huì)選取整個(gè)軸,。
array([ 1,  4,  7, 10])
arr[:,1:2]
array([[ 1],
       [ 4],
       [ 7],
       [10]])
arr[2:,1:]
array([[ 7,  8],
       [10, 11]])

2.3 數(shù)組的運(yùn)算

2.3.1 數(shù)組和標(biāo)量間的運(yùn)算

arr = np.array([1,2,3])
arr*10
array([10, 20, 30])
arr*arr
array([1, 4, 9])
arr-arr

array([0, 0, 0])
arr = np.random.randn(3,3)
arr

array([[-1.10434004, -0.38970982,  0.68926712],
       [-1.11139007, -0.60620091, -0.43321889],
       [-0.18284226,  1.65765572, -0.79486849]])
np.abs(arr)  # 通過abs函數(shù)求絕對(duì)值
array([[1.10434004, 0.38970982, 0.68926712],
       [1.11139007, 0.60620091, 0.43321889],
       [0.18284226, 1.65765572, 0.79486849]])
np.square(arr)  #square函數(shù)求平方

array([[1.21956692, 0.15187374, 0.47508917],
       [1.23518789, 0.36747954, 0.1876786 ],
       [0.03343129, 2.74782248, 0.63181591]])
arr1 = np.random.randint(1,10,size=(5))
arr1
array([9, 1, 8, 2, 3])
arr2 = np.random.randint(1,10,size = (5))
arr2
array([8, 8, 8, 1, 8])
np.add(arr1,arr2)  #add函數(shù)用于兩個(gè)數(shù)組的相加
array([17,  9, 16,  3, 11])
np.minimum(arr1,arr2) #minimum函數(shù)可以計(jì)算元素最小值
array([8, 1, 8, 1, 3])
arr = np.random.normal(2,4,size = (6))
arr

array([ 5.06802817, -1.26210681, -2.55346754, -5.70727512,  3.25494685,
        5.86244808])
np.modf(arr)
# 有些通用函數(shù)還可以返回兩個(gè)數(shù)組,，例如modf函數(shù)，可以返回?cái)?shù)組元素的小數(shù)和整數(shù)部分
(array([ 0.06802817, -0.26210681, -0.55346754, -0.70727512,  0.25494685,
         0.86244808]),
 array([ 5., -1., -2., -5.,  3.,  5.]))

2.3.3 條件邏輯運(yùn)算

arr1 = np.array([1,2,3,4])
arr2 = np.array([5,6,7,8])
cond = np.array([True,False,False,True])
result = np.where(cond,arr1,arr2)
result

#如果需要通過cond的值來選取arr1和arr2的值,，
#當(dāng)cond為True時(shí),，選擇arr1的值，否則選擇arr2的值
array([1, 6, 7, 4])
arr = np.random.randn(4,4)
arr

array([[-0.09650043,  0.31929566,  0.67190973, -0.951201  ],
       [ 1.10148109,  0.91289976,  1.24039204, -2.09494457],
       [-1.47448762, -1.61059505, -0.02424096, -1.00035527],
       [-1.71134107, -0.73639959, -0.63883441,  0.67764956]])
new_arr = np.where(arr>0,1,-1)
new_arr
array([[-1,  1,  1, -1],
       [ 1,  1,  1, -1],
       [-1, -1, -1, -1],
       [-1, -1, -1,  1]])

2.3.4 統(tǒng)計(jì)運(yùn)算

arr = np.random.randn(4,4)
arr

array([[-0.33407573,  1.89601251, -0.25068698,  1.09899693],
       [-0.7436235 ,  0.96724716,  2.51586733,  0.78559764],
       [-0.61260085, -0.1447176 ,  0.75132781,  0.91617282],
       [-0.33995589, -1.61785421, -0.12434649,  1.02720805]])
arr.sum()
5.790568998522431
arr.mean()
0.36191056240765196
arr.std()  #std函數(shù)用于求標(biāo)準(zhǔn)差
1.0283623213025257
arr

array([[-0.33407573,  1.89601251, -0.25068698,  1.09899693],
       [-0.7436235 ,  0.96724716,  2.51586733,  0.78559764],
       [-0.61260085, -0.1447176 ,  0.75132781,  0.91617282],
       [-0.33995589, -1.61785421, -0.12434649,  1.02720805]])
arr.mean(axis = 1)  #用于計(jì)算指定軸方向的統(tǒng)計(jì)值
array([ 0.60256168,  0.88127216,  0.22754555, -0.26373714])
arr.sum(0)
array([-2.03025597,  1.10068786,  2.89216167,  3.82797544])
arr = np.arange(9).reshape(3,3)
arr

array([[0, 1, 2],
       [3, 4, 5],
       [6, 7, 8]])
arr.cumsum(0)  #所有元素的累積和
array([[ 0,  1,  2],
       [ 3,  5,  7],
       [ 9, 12, 15]], dtype=int32)
arr.cumprod(1)  #所有元素的累積積
array([[  0,   0,   0],
       [  3,  12,  60],
       [  6,  42, 336]], dtype=int32)

2.3.5 布爾型數(shù)組運(yùn)算

arr = np.random.randn(20)
arr
array([-0.20661816,  0.10874043, -0.40256083, -0.19071145, -1.26373101,
        1.56890466, -1.01231556, -0.18248971,  1.23433954, -0.64327127,
       -1.64106717, -1.20622075,  0.96636457, -0.13955697,  1.31806702,
       -1.18310487,  0.02948617, -2.16793703,  0.88963934,  1.01751536])
arr = np.array([True,False,False,True])
arr

array([ True, False, False,  True])
arr.any()  #any方法用于測(cè)試數(shù)組中是否存在一個(gè)或多個(gè)True
True
arr.all()   #all方法用于檢查數(shù)組中的所有值是否為True

False

2.3.6 排序

arr = np.random.randn(10)
arr

array([-2.08464944,  1.79606612,  0.88671682, -1.5369521 , -1.46048203,
        1.18515803,  2.20130482, -0.32108926, -0.7320761 ,  0.12610864])
arr.sort()
arr
array([-2.08464944, -1.5369521 , -1.46048203, -0.7320761 , -0.32108926,
        0.12610864,  0.88671682,  1.18515803,  1.79606612,  2.20130482])
arr = np.random.randn(5,3)
arr
array([[-0.90307983, -0.54527699,  0.5234602 ],
       [ 0.01153178, -0.87114807, -0.07794886],
       [-0.34698081, -0.07907388,  0.60341693],
       [-0.06539896, -0.34155509,  1.37263374],
       [-0.5427262 , -0.09412761,  1.20774627]])
arr.sort(1)   #對(duì)于多維數(shù)組,，可以通過指定軸方向進(jìn)行排
arr

array([[-0.90307983, -0.54527699,  0.5234602 ],
       [-0.87114807, -0.07794886,  0.01153178],
       [-0.34698081, -0.07907388,  0.60341693],
       [-0.34155509, -0.06539896,  1.37263374],
       [-0.5427262 , -0.09412761,  1.20774627]])

2.4 數(shù)組的存取

arr = np.arange(12).reshape(3,4)
arr

array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
np.savetxt("che1ex1.csv",arr,fmt="%d",delimiter=",")
arr = np.loadtxt("che1ex1.csv",delimiter=",")
arr

array([[ 0.,  1.,  2.,  3.],
       [ 4.,  5.,  6.,  7.],
       [ 8.,  9., 10., 11.]])
from PIL import Image
im = np.array(Image.open("C:/Users/itcast/Desktop/image/1.jpg"))
print(im.shape,im.dtype)
(598, 601, 3) uint8
im  #通過圖中的代碼可以看出,，圖像轉(zhuǎn)換成三維數(shù)組后，維度分別為寬度,、長度和RGB值,。
array([[[128, 129, 131],
        [ 16,   4,   0],
        [246, 208,  61],
        ...,
        [255, 209,  10],
        [243, 208,  56],
        [ 11,   4,   0]],

       [[129, 130, 132],
        [ 14,   2,   0],
        [246, 208,  61],
        ...,
        [255, 209,  10],
        [243, 208,  56],
        [ 12,   5,   0]],

       [[123, 124, 126],
        [ 20,   8,   0],
        [246, 208,  61],
        ...,
        [255, 209,  10],
        [243, 208,  56],
        [ 11,   4,   0]],

       ...,

       [[139, 139, 137],
        [251, 243, 197],
        [248, 221,  90],
        ...,
        [255, 219,  51],
        [248, 217,  90],
        [  8,   5,   0]],

       [[120, 120, 118],
        [255, 255, 212],
        [243, 216,  85],
        ...,
        [255, 219,  51],
        [248, 217,  90],
        [255, 255, 216]],

       [[116, 116, 114],
        [255, 255, 212],
        [237, 210,  79],
        ...,
        [255, 220,  52],
        [249, 218,  91],
        [255, 255, 218]]], dtype=uint8)
b = [200,200,200] -im
new_im =  Image.fromarray(b.astype("uint8"))
new_im.save('C:/Users/itcast/Desktop/image/3.jpg')