Numpy用法总结

import numpy as npimport pandas as pdfrom pandas import Series, DataFrame

ndarray

一、创建ndarray

# 使用np.array()由Python list 创建data = [1,2,3,4]n1 = np.array(data)n1

array([1, 2, 3, 4])

# 使用np的routines函数创建# 创建多个都是1np.ones(5,dtype=int)

array([1, 1, 1, 1, 1])

# 创建多个 都是0np.zeros(5,dtype=int)

array([0, 0, 0, 0, 0])

# 生成指定shape形状的，值都为fill_valuenp.full(shape=(3,4),fill_value=1024)

array([[1024, 1024, 1024, 1024],       [1024, 1024, 1024, 1024],       [1024, 1024, 1024, 1024]])

# 生成对角线都为1，其余都为0，N为rows，M为columns，不指定，默认相等N=Mnp.eye(N=4,M=3)

array([[ 1.,  0.,  0.],       [ 0.,  1.,  0.],       [ 0.,  0.,  1.],       [ 0.,  0.,  0.]])

# 线性生成矩阵，计算start和stop，均分np.linspace(1,100,num=50,dtype=int)

array([  1,   3,   5,   7,   9,  11,  13,  15,  17,  19,  21,  23,  25,        27,  29,  31,  33,  35,  37,  39,  41,  43,  45,  47,  49,  51,        53,  55,  57,  59,  61,  63,  65,  67,  69,  71,  73,  75,  77,        79,  81,  83,  85,  87,  89,  91,  93,  95,  97, 100])

# 生成0,100之间，每次步数为5，左闭右开np.arange(0,100,step=5)

array([ 0,  5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80,       85, 90, 95])

# 生成随机整数，左闭右开，size为数量，可以指定形状np.random.randint(0,100,size=(4,3))

array([[70, 19, 93],       [24, 86, 92],       [77, 99, 88],       [18, 55, 29]])

# 生成标准正太分布np.random.randn(100)

array([-0.07352457, -1.3136685 ,  1.51471995, -0.08136916, -1.11574765,        0.17386269, -1.31727833, -0.41155414,  1.30430032, -0.86782806,       -0.76215743, -0.50927113,  1.52628811, -0.18189482, -0.02033947,        1.97168795,  0.24356417,  0.98391052, -0.41973603,  0.32262384,       -1.26531272,  0.51957893, -0.53221152, -0.18077624, -0.04793903,       -1.07987469, -0.14964406, -1.50501509,  0.22443113,  1.02252398,        1.8288228 ,  0.14265383, -1.07763155, -1.45768047,  0.57438741,       -0.02607326,  0.76216751, -0.21004165, -1.45035988, -2.21018297,        0.18629033,  0.51804587,  0.6108886 , -1.79317464, -0.01333462,       -1.02223808,  1.02241954, -0.21770695, -0.13238381, -2.43526884,        1.77094614,  1.58023618,  0.86634068,  0.63893764, -0.09904435,        0.19179536,  0.70664661,  0.18497984,  1.06407283,  0.08213135,        0.1189964 , -0.05590335, -0.73709569, -0.14316891, -0.90192259,        0.53208371,  1.52754737, -0.06674151,  1.07920869,  0.42858941,       -1.15588987, -1.10701065,  0.01149264, -1.36904717, -0.0157708 ,        0.81281272, -0.87613834, -0.05256806,  0.36815293, -1.19499011,        1.28316201,  0.68097169, -0.02886232, -0.45725581,  2.17369421,        0.67138163,  1.7039416 ,  0.16232207, -0.3319481 ,  0.8932495 ,       -1.67827068, -1.61199462,  0.60000361, -1.52476205, -0.28960245,        0.98590361, -0.67797509, -1.03187526, -0.84967747,  0.11290939])

# 生成以10为中心，偏移5，的10个数np.random.normal(10,5,10)

array([  8.41062812,   7.48295308,  17.81455213,  15.14675061,         3.74643964,  10.79360893,   5.2854533 ,  16.02048099,         3.22449646,   3.80641927])

# 生成0-1的随机数，左闭右开np.random.random(10)

array([ 0.68289218,  0.80123252,  0.91247595,  0.73536506,  0.867282  ,        0.89442711,  0.06441671,  0.80846567,  0.77142565,  0.19021568])

二、ndarray的属性

# ndim：维度；shape：形状；size：总长度；dtype：元素类型n = np.random.randint(100,size=(4,4))display(n.ndim, n.shape, n.size, n.dtype)

2(4, 4)16dtype('int64')

三、ndarray的基本操作

n = np.random.randint(100,size=(4,4,4))n

array([[[16, 66, 17, 90],        [97, 29, 56, 95],        [10, 97, 63, 22],        [39, 83, 25, 89]],       [[52, 88, 36, 20],        [64, 25, 51, 78],        [48, 95, 87, 63],        [97, 73, 83, 55]],       [[16, 16,  7, 51],        [10, 99, 83, 10],        [45, 35,  5, 20],        [68, 71, 23,  7]],       [[81, 20, 55, 63],        [ 8, 39,  9, 40],        [66,  6, 79, 92],        [10, 79, 94, 36]]])

# 索引# 取出特定的值，还可以进行修改n[0,0,0]

16

# 切片n[:]

array([[[16, 66, 17, 90],        [97, 29, 56, 95],        [10, 97, 63, 22],        [39, 83, 25, 89]],       [[52, 88, 36, 20],        [64, 25, 51, 78],        [48, 95, 87, 63],        [97, 73, 83, 55]],       [[16, 16,  7, 51],        [10, 99, 83, 10],        [45, 35,  5, 20],        [68, 71, 23,  7]],       [[81, 20, 55, 63],        [ 8, 39,  9, 40],        [66,  6, 79, 92],        [10, 79, 94, 36]]])

# 多维切片n[:,0:2]

array([[[16, 66, 17, 90],        [97, 29, 56, 95]],       [[52, 88, 36, 20],        [64, 25, 51, 78]],       [[16, 16,  7, 51],        [10, 99, 83, 10]],       [[81, 20, 55, 63],        [ 8, 39,  9, 40]]])

# 数据反转n[::-1]

array([[[81, 20, 55, 63],        [ 8, 39,  9, 40],        [66,  6, 79, 92],        [10, 79, 94, 36]],       [[16, 16,  7, 51],        [10, 99, 83, 10],        [45, 35,  5, 20],        [68, 71, 23,  7]],       [[52, 88, 36, 20],        [64, 25, 51, 78],        [48, 95, 87, 63],        [97, 73, 83, 55]],       [[16, 66, 17, 90],        [97, 29, 56, 95],        [10, 97, 63, 22],        [39, 83, 25, 89]]])

# 变形n = np.random.randint(100,size=16).reshape((4,4))n

array([[67, 11,  6, 41],       [51,  9, 30, 73],       [98, 75, 52, 84],       [34, 17,  2, 19]])

# 级联# 级联的参数是列表，要加中括号或者小括号# 维度必须相同# 级联的方向，默认是shape这个touple的第一个值所代表的方向的维度# 可通过axis参数改变级联的方向n1 = np.random.randint(100, size=(4,4))n2 = np.random.randint(100, size=(4,4))display(n1,n2)

array([[78,  0,  3,  0],       [28, 72, 92, 59],       [46, 14, 97, 63],       [15, 45,  2, 63]])array([[85, 61, 30, 21],       [59, 66, 66, 36],       [87, 72, 15, 72],       [11, 33, 12, 16]])

np.concatenate((n1,n2))

array([[78,  0,  3,  0],       [28, 72, 92, 59],       [46, 14, 97, 63],       [15, 45,  2, 63],       [85, 61, 30, 21],       [59, 66, 66, 36],       [87, 72, 15, 72],       [11, 33, 12, 16]])

np.concatenate((n1,n2),axis=1)

array([[78,  0,  3,  0, 85, 61, 30, 21],       [28, 72, 92, 59, 59, 66, 66, 36],       [46, 14, 97, 63, 87, 72, 15, 72],       [15, 45,  2, 63, 11, 33, 12, 16]])

# 水平级联与竖直级联# 处理自己，进行维度的改变# hstack水平级联，horizonn_stack = np.hstack(n1)n_stack

array([78,  0,  3,  0, 28, 72, 92, 59, 46, 14, 97, 63, 15, 45,  2, 63])

# vstack数值级联，verticalnp.vstack(n_stack)

array([[78],       [ 0],       [ 3],       [ 0],       [28],       [72],       [92],       [59],       [46],       [14],       [97],       [63],       [15],       [45],       [ 2],       [63]])

# 与级联类似，三个函数完成切分的工作# 按照索引，下标3(不包括)切一下，下标5(不包括)切一下>>>a = np.random.randint(0, 10, size=20)>>>np.split(a, [3,5])[array([3, 4, 4]),    array([4, 8]),    array([4, 6, 4, 9, 8, 7, 2, 1, 9, 6, 0, 4, 2, 8, 6]]

>>>b = np.random.randint(0,10,size = 20).reshape((4,5))>>>barray([[6, 0, 1, 0, 7],        [3, 3, 5, 9, 4],        [4, 9, 5, 4, 5],        [5, 6, 9, 5, 3]])

# 下标1(不包括1，其实是下标0)切一下，下标3(其实是2)切一下>>>np.vsplit(b,(1,3))[array([[6, 0, 1, 0, 7]]), array([[3, 3, 5, 9, 4],            [4, 9, 5, 4, 5]]), array([[5, 6, 9, 5, 3]])]

>>>np.hsplit(b, (2,3))[array([[6, 0],        [3, 3],        [4, 9],        [5, 6]]), array([[1],        [5],        [5],        [9]]), array([[0, 7],        [9, 4],        [4, 5],        [5, 3]])]

Function Name    NaN-safe Version    Descriptionnp.sum              np.nansum     Compute sum of elementsnp.prod             np.nanprod    Compute product of elementsnp.mean             np.nanmean    Compute mean of elementsnp.std              np.nanstd     Compute standard deviationnp.var              np.nanvar     Compute variancenp.min              np.nanmin     Find minimum valuenp.max              np.nanmax     Find maximum valuenp.argmin           np.nanargmin  Find index of minimum valuenp.argmax           np.nanargmax  Find index of maximum valuenp.median           np.nanmedian  Compute median of elementsnp.percentile       np.nanpercentile    Compute rank-based statistics of elementsnp.any                  N/A    Evaluate whether any elements are truenp.all                  N/A    Evaluate whether all elements are truenp.power               幂运算

# np.sum 和 np.nansum 的区别。nan： not a number>>>b = np.array([1, 2, None])>>>barray([1, 2, None], dtype=object)# np.nansum(b)会报错# np.nan == not a number>>>c = np.array([1,10,np.nan])>>>np.nansum()11.0

# 矩阵积>>>a = np.array([[-3,0],[5,0.5]])>>>b = np.array([[-7,2],[4,6]])>>>display(np.dot(a,b),a*b)array([[ 21.,  -6.],       [-33.,  13.]])array([[ 21.,   0.],       [ 20.,   3.]])

# np.sort()与ndarray.sort()都可以，但有区别：# np.sort()不改变输入# ndarray.sort()本地处理，不占用空间，但改变输入