100道练习带你玩转Numpy

BG6RSH · 帖子由 **BG6RSH** » 周四 7月 30, 2020 9:58 pm

Numpy是用Python做数据分析所必须要掌握的基础库之一，它可以用来存储和处理大型矩阵，并且Numpy提供了许多高级的数值编程工具，如：矩阵数据类型、矢量处理，以及精密的运算库，专为进行严格的数字处理而产生。

本文内容由科赛网翻译整理自Github开源项目(部分题目保留了原文作参考)，建议读者完成科赛网 Numpy快速上手指南 --- 基础篇和 Numpy快速上手指南 --- 进阶篇这两篇教程的学习之后，再对此教程进行调试学习。

以下为100道Numpy习题及答案

Code: [全选] [Expand/Collapse]

1. 导入numpy库并简写为 np
 
(提示: import … as …)
 
In [ ]:
 
# import numpy as np
 
2. 打印numpy的版本和配置说明
 
(提示: np.__version__, np.show_config)
 
In [ ]:
 
# print(np.__version__)
 
# np.show_config()
 
3. 创建一个长度为10的空向量
 
(提示: np.zeros)
 
In [ ]:
 
# Z = np.zeros(10)
 
# print(Z)
 
4. 如何找到任何一个数组的内存大小？
 
(提示: size, itemsize)
 
In [ ]:
 
# Z = np.zeros((10,10))
 
# print("%d bytes" % (Z.size * Z.itemsize))
 
5. 如何从命令行得到numpy中add函数的说明文档?
 
(提示: http://np.info)
 
In [ ]:
 
# http://numpy.info(numpy.add)
 
6. 创建一个长度为10并且除了第五个值为1的空向量
 
(提示: array[4])
 
In [ ]:
 
# Z = np.zeros(10)
 
# Z[4] = 1
 
# print(Z)
 
7. 创建一个值域范围从10到49的向量
 
(提示: np.arange)
 
In [ ]:
 
# Z = np.arange(10,50)
 
# print(Z)
 
8. 反转一个向量(第一个元素变为最后一个)
 
(提示: array[::-1])
 
In [ ]:
 
# Z = np.arange(50)
 
# Z = Z[::-1]
 
# print(Z)
 
9. 创建一个 3x3 并且值从0到8的矩阵
 
(提示: reshape)
 
In [ ]:
 
# Z = np.arange(9).reshape(3,3)
 
# print(Z)
 
10. 找到数组[1,2,0,0,4,0]中非0元素的位置索引
 
(提示: np.nonzero)
 
In [ ]:
 
# nz = np.nonzero([1,2,0,0,4,0])
 
# print(nz)
 
11. 创建一个 3x3 的单位矩阵
 
(提示: np.eye)
 
In [ ]:
 
# Z = np.eye(3)
 
# print(Z)
 
12. 创建一个 3x3x3的随机数组
 
(提示: np.random.random)
 
In [ ]:
 
# Z = np.random.random((3,3,3))
 
# print(Z)
 
13. 创建一个 10x10 的随机数组并找到它的最大值和最小值
 
(提示: min, max)
 
In [ ]:
 
# Z = np.random.random((10,10))
 
# Zmin, Zmax = Z.min(), Z.max()
 
# print(Zmin, Zmax)
 
14. 创建一个长度为30的随机向量并找到它的平均值
 
(提示: mean)
 
In [ ]:
 
# Z = np.random.random(30)
 
# m = Z.mean()
 
# print(m)
 
15.创建一二维数组，其中边界值为1，其余值为0
 
(提示: array[1:-1, 1:-1])
 
In [ ]:
 
# Z = np.ones((10,10))
 
# Z[1:-1,1:-1] = 0
 
# print(Z)
 
16. 对于一个存在在数组，如何添加一个用0填充的边界?
 
(提示: np.pad)
 
In [ ]:
 
# Z = np.ones((5,5))
 
# Z = np.pad(Z, pad_width=1, mode='constant', constant_values=0)
 
# print(Z)
 
17. 以下表达式运行的结果分别是什么?
 
(提示: NaN = not a number, inf = infinity)
 
0*np.nan
 
np.nan==np.nan
 
np.inf>np.nan
 
np.nan-np.nan
 
0.3==3*0.1
 
In [ ]:
 
# print(0 * np.nan)
 
In [ ]:
 
# print(np.nan == np.nan)
 
In [ ]:
 
# print(np.inf > np.nan)
 
In [ ]:
 
# print(np.nan - np.nan)
 
In [ ]:
 
# print(0.3 == 3 * 0.1)
 
18. 创建一个 5x5的矩阵，并设置值1,2,3,4落在其对角线下方位置
 
(提示: np.diag)
 
In [ ]:
 
# Z = np.diag(1+np.arange(4),k=-1)
 
# print(Z)
 
19. 创建一个8x8 的矩阵，并且设置成棋盘样式
 
(提示: array[::2])
 
In [ ]:
 
# Z = np.zeros((8,8),dtype=int)
 
#Z[1::2,::2] = 1
 
# Z[::2,1::2] = 1
 
# print(Z)
 
20. 考虑一个 (6,7,8) 形状的数组，其第100个元素的索引(x,y,z)是什么?
 
(提示: np.unravel_index)
 
In [ ]:
 
# print(np.unravel_index(100,(6,7,8)))
 
21. 用tile函数去创建一个 8x8的棋盘样式矩阵
 
(提示: np.tile)
 
In [ ]:
 
# Z = np.tile( np.array([[0,1],[1,0]]), (4,4))
 
# print(Z)
 
22. 对一个5x5的随机矩阵做归一化
 
(提示: (x - min) / (max - min))
 
In [ ]:
 
# Z = np.random.random((5,5))
 
# Zmax, Zmin = Z.max(), Z.min()
 
# Z = (Z - Zmin)/(Zmax - Zmin)
 
# print(Z)
 
23. 创建一个将颜色描述为(RGBA)四个无符号字节的自定义dtype？
 
(提示: np.dtype)
 
In [ ]:
 
# color = np.dtype([("r", np.ubyte, 1),
 
# ("g", np.ubyte, 1),
 
# ("b", np.ubyte, 1),
 
# ("a", np.ubyte, 1)])
 
# color
 
24. 一个5x3的矩阵与一个3x2的矩阵相乘，实矩阵乘积是什么？
 
(提示: np.dot | @)
 
In [ ]:
 
# Z = np.dot(np.ones((5,3)), np.ones((3,2)))
 
# print(Z)
 
25. 给定一个一维数组，对其在3到8之间的所有元素取反
 
(提示: >, <=)
 
In [ ]:
 
# Z = np.arange(11)
 
# Z[(3 < Z) & (Z <= 8)] *= -1
 
# print(Z)
 
26. 下面脚本运行后的结果是什么?
 
(提示: np.sum)
 
In [ ]:
 
# print(sum(range(5),-1))
 
In [ ]:
 
# from numpy import *
 
# print(sum(range(5),-1))
 
27. 考虑一个整数向量Z,下列表达合法的是哪个?
 
Z**Z
 
2<<Z>>2
 
Z<-Z
 
1j*Z
 
Z/1/1
 
Z<Z>Z
 
In [ ]:
 
# Z = np.arange(5)
 
# Z ** Z # legal
 
In [ ]:
 
# Z = np.arange(5)
 
# 2 << Z >> 2 # false
 
In [ ]:
 
# Z = np.arange(5)
 
# Z <- Z # legal
 
In [ ]:
 
# Z = np.arange(5)
 
# 1j*Z # legal
 
In [ ]:
 
# Z = np.arange(5)
 
# Z/1/1 # legal
 
In [ ]:
 
# Z = np.arange(5)
 
# Z<Z>Z # false
 
28. 下列表达式的结果分别是什么?
 
np.array(0) /np.array(0)
 
np.array(0) //np.array(0)
 
np.array([np.nan]).astype(int).astype(float)
 
In [ ]:
 
# print(np.array(0) / np.array(0))
 
In [ ]:
 
# print(np.array(0) // np.array(0))
 
In [ ]:
 
# print(np.array([np.nan]).astype(int).astype(float))
 
29. 如何从零位对浮点数组做舍入 ?
 
(提示: np.uniform, np.copysign, np.ceil, np.abs)
 
In [ ]:
 
# Z = np.random.uniform(-10,+10,10)
 
# print (np.copysign(np.ceil(np.abs(Z)), Z))
 
30. 如何找到两个数组中的共同元素?
 
(提示: np.intersect1d)
 
In [ ]:
 
# Z1 = np.random.randint(0,10,10)
 
# Z2 = np.random.randint(0,10,10)
 
# print(np.intersect1d(Z1,Z2))
 
31. 如何忽略所有的 numpy 警告(尽管不建议这么做)?
 
(提示: np.seterr, np.errstate)
 
# Suicide mode on
 
defaults=np.seterr(all="ignore")
 
Z=np.ones(1) /0
 
# Back to sanity
 
_=np.seterr(**defaults)
 
Anequivalentway, withacontextmanager:
 
withnp.errstate(divide='ignore'):
 
Z=np.ones(1) /0
 
32. 下面的表达式是正确的吗?
 
(提示: imaginary number)
 
np.sqrt(-1) ==np.emath.sqrt(-1)
 
In [ ]:
 
# np.sqrt(-1) == np.emath.sqrt(-1) # False
 
33. 如何得到昨天，今天，明天的日期?
 
(提示: np.datetime64, np.timedelta64)
 
In [ ]:
 
# yesterday = np.datetime64('today', 'D') - np.timedelta64(1, 'D')
 
# today = np.datetime64('today', 'D')
 
# tomorrow = np.datetime64('today', 'D') + np.timedelta64(1, 'D')
 
# print ("Yesterday is " + str(yesterday))
 
# print ("Today is " + str(today))
 
# print ("Tomorrow is "+ str(tomorrow))
 
34. 如何得到所有与2016年7月对应的日期？
 
(提示: np.arange(dtype=datetime64['D']))
 
In [ ]:
 
# Z = np.arange('2016-07', '2016-08', dtype='datetime64[D]')
 
# print(Z)
 
35. 如何直接在位计算(A+B)\*(-A/2)(不建立副本)?
 
(提示: np.add(out=), np.negative(out=), np.multiply(out=), np.divide(out=))
 
In [ ]:
 
# A = np.ones(3)*1
 
# B = np.ones(3)*2
 
# C = np.ones(3)*3
 
# np.add(A,B,out=B)
 
In [ ]:
 
# np.divide(A,2,out=A)
 
In [ ]:
 
# np.negative(A,out=A)
 
In [ ]:
 
# np.multiply(A,B,out=A)
 
36. 用五种不同的方法去提取一个随机数组的整数部分
 
(提示: %, np.floor, np.ceil, astype, np.trunc)
 
In [ ]:
 
# Z = np.random.uniform(0,10,10)
 
# print (Z - Z%1)
 
In [ ]:
 
# print (np.floor(Z))
 
In [ ]:
 
# print (np.ceil(Z)-1)
 
In [ ]:
 
# print (Z.astype(int))
 
In [ ]:
 
# print (np.trunc(Z))
 
37. 创建一个5x5的矩阵，其中每行的数值范围从0到4
 
(提示: np.arange)
 
In [ ]:
 
# Z = np.zeros((5,5))
 
# Z += np.arange(5)
 
# print (Z)
 
38. 通过考虑一个可生成10个整数的函数，来构建一个数组
 
(提示: np.fromiter)
 
In [ ]:
 
# def generate():
 
# for x in range(10):
 
# yield x
 
# Z = np.fromiter(generate(),dtype=float,count=-1)
 
# print (Z)
 
39. 创建一个长度为10的随机向量，其值域范围从0到1，但是不包括0和1
 
(提示: np.linspace)
 
In [ ]:
 
# Z = np.linspace(0,1,11,endpoint=False)[1:]
 
# print (Z)
 
40. 创建一个长度为10的随机向量，并将其排序
 
(提示: sort)
 
In [ ]:
 
# Z = np.random.random(10)
 
# Z.sort()
 
# print (Z)
 
41.对于一个小数组，如何用比 np.sum更快的方式对其求和？
 
(提示: np.add.reduce)
 
In [ ]:
 
# Z = np.arange(10)
 
# np.add.reduce(Z)
 
42. 对于两个随机数组A和B，检查它们是否相等
 
(提示: np.allclose, np.array_equal)
 
In [ ]:
 
# A = np.random.randint(0,2,5)
 
# B = np.random.randint(0,2,5)
 
# # Assuming identical shape of the arrays and a tolerance for the comparison of values
 
# equal = np.allclose(A,B)
 
# print(equal)
 
In [ ]:
 
# # 方法2
 
# # Checking both the shape and the element values, no tolerance (values have to be exactly equal)
 
# equal = np.array_equal(A,B)
 
# print(equal)
 
43. 创建一个只读数组(read-only)
 
(提示: flags.writeable)
 
# 使用如下过程实现
 
Z=np.zeros(10)
 
Z.flags.writeable=False
 
Z[0] =1
 
---------------------------------------------------------------------------
 
ValueErrorTraceback(mostrecentcalllast)
 
<ipython-input-54-6fd4c6570dd1>in<module>()
 
1Z=np.zeros(10)
 
2Z.flags.writeable=False
 
---->3Z[0] =1
 
ValueError: assignmentdestinationisread-only
 
44. 将笛卡尔坐标下的一个10x2的矩阵转换为极坐标形式
 
(hint: np.sqrt, np.arctan2)
 
In [ ]:
 
# Z = np.random.random((10,2))
 
# X,Y = Z[:,0], Z[:,1]
 
# R = np.sqrt(X**2+Y**2)
 
# T = np.arctan2(Y,X)
 
# print (R)
 
# print (T)
 
45. 创建一个长度为10的向量，并将向量中最大值替换为1
 
(提示: argmax)
 
In [ ]:
 
# Z = np.random.random(10)
 
# Z[Z.argmax()] = 0
 
# print (Z)
 
46. 创建一个结构化数组，并实现 x 和 y 坐标覆盖 [0,1]x[0,1] 区域
 
(提示: np.meshgrid)
 
In [ ]:
 
# Z = np.zeros((5,5), [('x',float),('y',float)])
 
# Z['x'], Z['y'] = np.meshgrid(np.linspace(0,1,5),
 
# np.linspace(0,1,5))
 
# print(Z)
 
47. 给定两个数组X和Y，构造Cauchy矩阵C (Cij =1/(xi - yj))
 
(提示: np.subtract.outer)
 
In [ ]:
 
# X = np.arange(8)
 
# Y = X + 0.5
 
# C = 1.0 / np.subtract.outer(X, Y)
 
# print(np.linalg.det(C))
 
48. 打印每个numpy标量类型的最小值和最大值？
 
(提示: np.iinfo, np.finfo, eps)
 
In [ ]:
 
# for dtype in [np.int8, np.int32, np.int64]:
 
# print(np.iinfo(dtype).min)
 
# print(np.iinfo(dtype).max)
 
# for dtype in [np.float32, np.float64]:
 
# print(np.finfo(dtype).min)
 
# print(np.finfo(dtype).max)
 
# print(np.finfo(dtype).eps)
 
49. 如何打印一个数组中的所有数值?
 
(提示: np.set_printoptions)
 
In [ ]:
 
# np.set_printoptions(threshold=np.nan)
 
# Z = np.zeros((16,16))
 
# print (Z)
 
50. 给定标量时，如何找到数组中最接近标量的值？
 
(提示: argmin)
 
In [ ]:
 
# Z = np.arange(100)
 
# v = np.random.uniform(0,100)
 
# index = (np.abs(Z-v)).argmin()
 
# print (Z[index])
 
51. 创建一个表示位置(x,y)和颜色(r,g,b)的结构化数组
 
(提示: dtype)
 
In [ ]:
 
# Z = np.zeros(10, [ ('position', [ ('x', float, 1),
 
# ('y', float, 1)]),
 
# ('color', [ ('r', float, 1),
 
# ('g', float, 1),
 
# ('b', float, 1)])])
 
# print (Z)
 
52. 对一个表示坐标形状为(100,2)的随机向量，找到点与点的距离
 
(提示: np.atleast_2d, T, np.sqrt)
 
In [ ]:
 
# Z = np.random.random((10,2))
 
# X,Y = np.atleast_2d(Z[:,0], Z[:,1])
 
# D = np.sqrt( (X-X.T)**2 + (Y-Y.T)**2)
 
# print (D)
 
In [ ]:
 
# # 方法2
 
# # Much faster with scipy
 
# import scipy
 
# # Thanks Gavin Heverly-Coulson (#issue 1)
 
# import scipy.spatial
 
# D = scipy.spatial.distance.cdist(Z,Z)
 
# print (D)
 
53. 如何将32位的浮点数(float)转换为对应的整数(integer)?
 
(提示: astype(copy=False))
 
In [ ]:
 
# Z = np.arange(10, dtype=np.int32)
 
# Z = Z.astype(np.float32, copy=False)
 
# print (Z)
 
54. 如何读取以下文件?
 
(提示: np.genfromtxt)
 
1, 2, 3, 4, 5
 
6, , , 7, 8
 
, , 9,10,11
 
参考链接
 
55. 对于numpy数组，enumerate的等价操作是什么？
 
(提示: np.ndenumerate, np.ndindex)
 
In [ ]:
 
# Z = np.arange(9).reshape(3,3)
 
# for index, value in np.ndenumerate(Z):
 
# print (index, value)
 
# for index in np.ndindex(Z.shape):
 
# print (index, Z[index])
 
56. 生成一个通用的二维Gaussian-like数组
 
(提示: np.meshgrid, np.exp)
 
In [ ]:
 
# X, Y = np.meshgrid(np.linspace(-1,1,10), np.linspace(-1,1,10))
 
# D = np.sqrt(X*X+Y*Y)
 
# sigma, mu = 1.0, 0.0
 
# G = np.exp(-( (D-mu)**2 / ( 2.0 * sigma**2 ) ) )
 
# print (G)
 
57. 对一个二维数组，如何在其内部随机放置p个元素?
 
(提示: np.put, np.random.choice)
 
In [ ]:
 
# n = 10
 
# p = 3
 
# Z = np.zeros((n,n))
 
# np.put(Z, np.random.choice(range(n*n), p, replace=False),1)
 
# print (Z)
 
58. 减去一个矩阵中的每一行的平均值
 
(提示: mean(axis=,keepdims=))
 
In [ ]:
 
# X = np.random.rand(5, 10)
 
# # Recent versions of numpy
 
# Y = X - X.mean(axis=1, keepdims=True)
 
# print(Y)
 
In [ ]:
 
# # 方法2
 
# # Older versions of numpy
 
# Y = X - X.mean(axis=1).reshape(-1, 1)
 
# print (Y)
 
59. 如何通过第n列对一个数组进行排序?
 
(提示: argsort)
 
In [ ]:
 
# Z = np.random.randint(0,10,(3,3))
 
# print (Z)
 
# print (Z[Z[:,1].argsort()])
 
60. 如何检查一个二维数组是否有空列？
 
(提示: any, ~)
 
In [ ]:
 
# Z = np.random.randint(0,3,(3,10))
 
# print ((~Z.any(axis=0)).any())
 
61. 从数组中的给定值中找出最近的值
 
(提示: np.abs, argmin, flat)
 
In [ ]:
 
# Z = np.random.uniform(0,1,10)
 
# z = 0.5
 
# m = Z.flat[np.abs(Z - z).argmin()]
 
# print (m)
 
62. 如何用迭代器(iterator)计算两个分别具有形状(1,3)和(3,1)的数组?
 
(提示: np.nditer)
 
In [ ]:
 
# A = np.arange(3).reshape(3,1)
 
# B = np.arange(3).reshape(1,3)
 
# it = np.nditer([A,B,None])
 
# for x,y,z in it:
 
# z[...] = x + y
 
# print (it.operands[2])
 
63. 创建一个具有name属性的数组类
 
(提示: class方法)
 
In [ ]:
 
# class NamedArray(np.ndarray):
 
# def __new__(cls, array, name="no name"):
 
# obj = np.asarray(array).view(cls)
 
# obj.name = name
 
# return obj
 
# def __array_finalize__(self, obj):
 
# if obj is None: return
 
# http://self.info = getattr(obj, 'name', "no name")
 
# Z = NamedArray(np.arange(10), "range_10")
 
# print (Z.name)
 
64. 考虑一个给定的向量，如何对由第二个向量索引的每个元素加1(小心重复的索引)?
 
(提示: np.bincount | np.add.at)
 
In [ ]:
 
# Z = np.ones(10)
 
# I = np.random.randint(0,len(Z),20)
 
# Z += np.bincount(I, minlength=len(Z))
 
# print(Z)
 
In [ ]:
 
# # 方法2
 
# np.add.at(Z, I, 1)
 
# print(Z)
 
65. 根据索引列表(I)，如何将向量(X)的元素累加到数组(F)?
 
(提示: np.bincount)
 
In [ ]:
 
# X = [1,2,3,4,5,6]
 
# I = [1,3,9,3,4,1]
 
# F = np.bincount(I,X)
 
# print (F)
 
66. 考虑一个(dtype=ubyte) 的 (w,h,3)图像，计算其唯一颜色的数量
 
(提示: np.unique)
 
In [ ]:
 
# w,h = 16,16
 
# I = np.random.randint(0,2,(h,w,3)).astype(np.ubyte)
 
# #Note that we should compute 256*256 first.
 
# #Otherwise numpy will only promote F.dtype to 'uint16' and overfolw will occur
 
# F = I[...,0]*(256*256) + I[...,1]*256 +I[...,2]
 
# n = len(np.unique(F))
 
# print (n)
 
67. 考虑一个四维数组，如何一次性计算出最后两个轴(axis)的和？
 
(提示: sum(axis=(-2,-1)))
 
In [ ]:
 
# A = np.random.randint(0,10,(3,4,3,4))
 
# # solution by passing a tuple of axes (introduced in numpy 1.7.0)
 
# sum = A.sum(axis=(-2,-1))
 
# print (sum)
 
In [ ]:
 
# # 方法2
 
# sum = A.reshape(A.shape[:-2] + (-1,)).sum(axis=-1)
 
# print (sum)
 
68. 考虑一个一维向量D，如何使用相同大小的向量S来计算D子集的均值？
 
(提示: np.bincount)
 
In [ ]:
 
# D = np.random.uniform(0,1,100)
 
# S = np.random.randint(0,10,100)
 
# D_sums = np.bincount(S, weights=D)
 
# D_counts = np.bincount(S)
 
# D_means = D_sums / D_counts
 
# print (D_means)
 
In [ ]:
 
# # 方法2
 
# import pandas as pd
 
# print(pd.Series(D).groupby(S).mean())
 
69. 如何获得点积 dot prodcut的对角线?
 
(提示: np.diag)
 
In [ ]:
 
# A = np.random.uniform(0,1,(5,5))
 
# B = np.random.uniform(0,1,(5,5))
 
# # slow version
 
# np.diag(np.dot(A, B))
 
In [ ]:
 
## 方法2
 
# # Fast version
 
# np.sum(A * B.T, axis=1)
 
In [ ]:
 
## 方法3
 
# # Faster version
 
# np.einsum("ij,ji->i", A, B)
 
70. 考虑一个向量[1,2,3,4,5],如何建立一个新的向量，在这个新向量中每个值之间有3个连续的零？
 
(提示: array[::4])
 
In [ ]:
 
# Z = np.array([1,2,3,4,5])
 
# nz = 3
 
# Z0 = np.zeros(len(Z) + (len(Z)-1)*(nz))
 
# Z0[::nz+1] = Z
 
# print (Z0)
 
71. 考虑一个维度(5,5,3)的数组，如何将其与一个(5,5)的数组相乘？
 
(提示: array[:, :, None])
 
In [ ]:
 
# A = np.ones((5,5,3))
 
# B = 2*np.ones((5,5))
 
# print (A * B[:,:,None])
 
72. 如何对一个数组中任意两行做交换?
 
(提示: array[[]] = array[[]])
 
In [ ]:
 
# A = np.arange(25).reshape(5,5)
 
# A[[0,1]] = A[[1,0]]
 
# print (A)
 
73. 考虑一个可以描述10个三角形的triplets，找到可以分割全部三角形的line segment
 
(提示: repeat, np.roll, np.sort, view, np.unique)
 
In [ ]:
 
# faces = np.random.randint(0,100,(10,3))
 
# F = np.roll(faces.repeat(2,axis=1),-1,axis=1)
 
# F = F.reshape(len(F)*3,2)
 
# F = np.sort(F,axis=1)
 
# G = F.view( dtype=[('p0',F.dtype),('p1',F.dtype)] )
 
# G = np.unique(G)
 
# print (G)
 
74. 给定一个二进制的数组C，如何产生一个数组A满足np.bincount(A)==C
 
(提示: np.repeat)
 
In [ ]:
 
# C = np.bincount([1,1,2,3,4,4,6])
 
# A = np.repeat(np.arange(len(C)), C)
 
# print (A)
 
75. 如何通过滑动窗口计算一个数组的平均数?
 
(提示: np.cumsum)
 
In [ ]:
 
# def moving_average(a, n=3) :
 
# ret = np.cumsum(a, dtype=float)
 
# ret[n:] = ret[n:] - ret[:-n]
 
# return ret[n - 1:] / n
 
# Z = np.arange(20)
 
# print(moving_average(Z, n=3))
 
76. 考虑一维数组Z，构建一个二维数组，其第一行是（Z [0]，Z [1]，Z [2]），每个后续行移1（最后一行应该是（ Z [-3]，Z [-2]，Z [-1]）
 
(提示: from numpy.lib import stride_tricks)
 
In [ ]:
 
# from numpy.lib import stride_tricks
 
# def rolling(a, window):
 
# shape = (a.size - window + 1, window)
 
# strides = (a.itemsize, a.itemsize)
 
# return stride_tricks.as_strided(a, shape=shape, strides=strides)
 
# Z = rolling(np.arange(10), 3)
 
# print (Z)
 
77. 如何对布尔值取反，或者原位(in-place)改变浮点数的符号(sign)？
 
(提示: np.logical_not, np.negative)
 
In [ ]:
 
# Z = np.random.randint(0,2,100)
 
# np.logical_not(Z, out=Z)
 
In [ ]:
 
# Z = np.random.uniform(-1.0,1.0,100)
 
# np.negative(Z, out=Z)
 
78. 考虑两组点集P0和P1去描述一组线(二维)和一个点p,如何计算点p到每一条线 i (P0[i],P1[i])的距离？
 
In [ ]:
 
# def distance(P0, P1, p):
 
# T = P1 - P0
 
# L = (T**2).sum(axis=1)
 
# U = -((P0[:,0]-p[...,0])*T[:,0] + (P0[:,1]-p[...,1])*T[:,1]) / L
 
# U = U.reshape(len(U),1)
 
# D = P0 + U*T - p
 
# return np.sqrt((D**2).sum(axis=1))
 
# P0 = np.random.uniform(-10,10,(10,2))
 
# P1 = np.random.uniform(-10,10,(10,2))
 
# p = np.random.uniform(-10,10,( 1,2))
 
# print (distance(P0, P1, p))
 
79.考虑两组点集P0和P1去描述一组线(二维)和一组点集P，如何计算每一个点 j(P[j]) 到每一条线 i (P0[i],P1[i])的距离？
 
In [ ]:
 
# # based on distance function from previous question
 
# P0 = np.random.uniform(-10, 10, (10,2))
 
# P1 = np.random.uniform(-10,10,(10,2))
 
# p = np.random.uniform(-10, 10, (10,2))
 
# print (np.array([distance(P0,P1,p_i) for p_i in p]))
 
80.考虑一个任意数组，写一个函数，提取一个固定形状的子部分，并以给定元素为中心（fill必要时填充一个值）
 
(提示: minimum, maximum)
 
In [ ]:
 
# Z = np.random.randint(0,10,(10,10))
 
# shape = (5,5)
 
# fill = 0
 
# position = (1,1)
 
# R = np.ones(shape, dtype=Z.dtype)*fill
 
# P = np.array(list(position)).astype(int)
 
# Rs = np.array(list(R.shape)).astype(int)
 
# Zs = np.array(list(Z.shape)).astype(int)
 
# R_start = np.zeros((len(shape),)).astype(int)
 
# R_stop = np.array(list(shape)).astype(int)
 
# Z_start = (P-Rs//2)
 
# Z_stop = (P+Rs//2)+Rs%2
 
# R_start = (R_start - np.minimum(Z_start,0)).tolist()
 
# Z_start = (np.maximum(Z_start,0)).tolist()
 
# R_stop = np.maximum(R_start, (R_stop - np.maximum(Z_stop-Zs,0))).tolist()
 
# Z_stop = (np.minimum(Z_stop,Zs)).tolist()
 
# r = [slice(start,stop) for start,stop in zip(R_start,R_stop)]
 
# z = [slice(start,stop) for start,stop in zip(Z_start,Z_stop)]
 
# R[r] = Z[z]
 
# print (Z)
 
# print (R)
 
81. 考虑一个数组Z = [1,2,3,4,5,6,7,8,9,10,11,12,13,14],如何生成一个数组R = [[1,2,3,4], [2,3,4,5], [3,4,5,6], ...,[11,12,13,14]]?
 
(提示: stride_tricks.as_strided)
 
In [ ]:
 
# Z = np.arange(1,15,dtype=np.uint32)
 
# R = stride_tricks.as_strided(Z,(11,4),(4,4))
 
# print (R)
 
82. 计算一个矩阵的秩
 
(提示: np.linalg.svd)
 
In [ ]:
 
# Z = np.random.uniform(0,1,(10,10))
 
# U, S, V = np.linalg.svd(Z) # Singular Value Decomposition
 
# rank = np.sum(S > 1e-10)
 
# print (rank)
 
83. 如何找到一个数组中出现频率最高的值？
 
(提示: np.bincount, argmax)
 
In [ ]:
 
# Z = np.random.randint(0,10,50)
 
# print (np.bincount(Z).argmax())
 
84. 从一个10x10的矩阵中提取出连续的3x3区块
 
(提示: stride_tricks.as_strided)
 
In [ ]:
 
# Z = np.random.randint(0,5,(10,10))
 
# n = 3
 
# i = 1 + (Z.shape[0]-3)
 
# j = 1 + (Z.shape[1]-3)
 
# C = stride_tricks.as_strided(Z, shape=(i, j, n, n), strides=Z.strides + Z.strides)
 
# print (C)
 
85. 创建一个满足 Z[i,j] == Z[j,i]的子类
 
(提示: class 方法)
 
In [ ]:
 
# class Symetric(np.ndarray):
 
# def __setitem__(self, index, value):
 
# i,j = index
 
# super(Symetric, self).__setitem__((i,j), value)
 
# super(Symetric, self).__setitem__((j,i), value)
 
# def symetric(Z):
 
# return np.asarray(Z + Z.T - np.diag(Z.diagonal())).view(Symetric)
 
# S = symetric(np.random.randint(0,10,(5,5)))
 
# S[2,3] = 42
 
# print (S)
 
86. 考虑p个 nxn 矩阵和一组形状为(n,1)的向量，如何直接计算p个矩阵的乘积(n,1)？
 
(提示: np.tensordot)
 
In [ ]:
 
# p, n = 10, 20
 
# M = np.ones((p,n,n))
 
# V = np.ones((p,n,1))
 
# S = np.tensordot(M, V, axes=[[0, 2], [0, 1]])
 
# print (S)
 
# It works, because:
 
# M is (p,n,n)
 
# V is (p,n,1)
 
# Thus, summing over the paired axes 0 and 0 (of M and V independently),
 
# and 2 and 1, to remain with a (n,1) vector.
 
87. 对于一个16x16的数组，如何得到一个区域(block-sum)的和(区域大小为4x4)?
 
(提示: np.add.reduceat)
 
In [ ]:
 
# Z = np.ones((16,16))
 
# k = 4
 
# S = np.add.reduceat(np.add.reduceat(Z, np.arange(0, Z.shape[0], k), axis=0),
 
# np.arange(0, Z.shape[1], k), axis=1)
 
# print (S)
 
88. 如何利用numpy数组实现Game of Life?
 
(提示: Game of Life)
 
In [ ]:
 
# def iterate(Z):
 
# # Count neighbours
 
# N = (Z[0:-2,0:-2] + Z[0:-2,1:-1] + Z[0:-2,2:] +
 
# Z[1:-1,0:-2] + Z[1:-1,2:] +
 
# Z[2: ,0:-2] + Z[2: ,1:-1] + Z[2: ,2:])
 
# # Apply rules
 
# birth = (N==3) & (Z[1:-1,1:-1]==0)
 
# survive = ((N==2) | (N==3)) & (Z[1:-1,1:-1]==1)
 
# Z[...] = 0
 
# Z[1:-1,1:-1][birth | survive] = 1
 
# return Z
 
# Z = np.random.randint(0,2,(50,50))
 
# for i in range(100): Z = iterate(Z)
 
# print (Z)
 
89. 如何找到一个数组的第n个最大值?
 
(提示: np.argsort | np.argpartition)
 
In [ ]:
 
# Z = np.arange(10000)
 
# np.random.shuffle(Z)
 
# n = 5
 
# # Slow
 
# print (Z[np.argsort(Z)[-n:]])
 
In [ ]:
 
# # 方法2
 
# # Fast
 
# print (Z[np.argpartition(-Z,n)[:n]])
 
90. 给定任意个数向量，创建笛卡尔积(每一个元素的每一种组合)
 
(提示: np.indices)
 
In [ ]:
 
# def cartesian(arrays):
 
# arrays = [np.asarray(a) for a in arrays]
 
# shape = (len(x) for x in arrays)
 
# ix = np.indices(shape, dtype=int)
 
# ix = ix.reshape(len(arrays), -1).T
 
# for n, arr in enumerate(arrays):
 
# ix[:, n] = arrays[n][ix[:, n]]
 
# return ix
 
# print (cartesian(([1, 2, 3], [4, 5], [6, 7])))
 
91. 如何从一个正常数组创建记录数组(record array)?
 
(提示: np.core.records.fromarrays)
 
In [ ]:
 
# Z = np.array([("Hello", 2.5, 3),
 
# ("World", 3.6, 2)])
 
# R = np.core.records.fromarrays(Z.T,
 
# names='col1, col2, col3',
 
# formats = 'S8, f8, i8')
 
# print (R)
 
92. 考虑一个大向量Z, 用三种不同的方法计算它的立方
 
(提示: np.power, \*, np.einsum)
 
In [ ]:
 
# x = np.random.rand()
 
# np.power(x,3)
 
In [ ]:
 
## 方法2
 
# x*x*x
 
In [ ]:
 
## 方法3
 
# np.einsum('i,i,i->i',x,x,x)
 
93. 考虑两个形状分别为(8,3) 和(2,2)的数组A和B. 如何在数组A中找到满足包含B中元素的行？(不考虑B中每行元素顺序)？
 
(提示: np.where)
 
In [ ]:
 
# A = np.random.randint(0,5,(8,3))
 
# B = np.random.randint(0,5,(2,2))
 
# C = (A[..., np.newaxis, np.newaxis] == B)
 
# rows = np.where(C.any((3,1)).all(1))[0]
 
# print (rows)
 
94. 考虑一个10x3的矩阵，分解出有不全相同值的行 (如 [2,2,3])
 
In [ ]:
 
# Z = np.random.randint(0,5,(10,3))
 
# print (Z)
 
# # solution for arrays of all dtypes (including string arrays and record arrays)
 
# E = np.all(Z[:,1:] == Z[:,:-1], axis=1)
 
# U = Z[~E]
 
# print (U)
 
In [ ]:
 
# # 方法2
 
# # soluiton for numerical arrays only, will work for any number of columns in Z
 
# U = Z[Z.max(axis=1) != Z.min(axis=1),:]
 
# print (U)
 
95. 将一个整数向量转换为matrix binary的表现形式
 
(提示: np.unpackbits)
 
In [ ]:
 
# I = np.array([0, 1, 2, 3, 15, 16, 32, 64, 128])
 
# B = ((I.reshape(-1,1) & (2**np.arange(8))) != 0).astype(int)
 
# print(B[:,::-1])
 
In [ ]:
 
# # 方法2
 
# print (np.unpackbits(I[:, np.newaxis], axis=1))
 
96. 给定一个二维数组，如何提取出唯一的(unique)行？
 
(提示: np.ascontiguousarray)
 
In [ ]:
 
# Z = np.random.randint(0,2,(6,3))
 
# T = np.ascontiguousarray(Z).view(np.dtype((np.void, Z.dtype.itemsize * Z.shape[1])))
 
# _, idx = np.unique(T, return_index=True)
 
# uZ = Z[idx]
 
# print (uZ)
 
97. 考虑两个向量A和B，写出用einsum等式对应的inner, outer, sum, mul函数
 
(提示: np.einsum)
 
In [ ]:
 
# A = np.random.uniform(0,1,10)
 
# B = np.random.uniform(0,1,10)
 
# print ('sum')
 
# print (np.einsum('i->', A))# np.sum(A)
 
In [ ]:
 
# print ('A * B')
 
# print (np.einsum('i,i->i', A, B)) # A * B
 
In [ ]:
 
# print ('inner')
 
# print (np.einsum('i,i', A, B)) # np.inner(A, B)
 
In [ ]:
 
# print ('outer')
 
# print (np.einsum('i,j->ij', A, B)) # np.outer(A, B)
 
98. 考虑一个由两个向量描述的路径(X,Y)，如何用等距样例(equidistant samples)对其进行采样(sample)?
 
(提示: np.cumsum, np.interp)
 
In [ ]:
 
# phi = np.arange(0, 10*np.pi, 0.1)
 
# a = 1
 
# x = a*phi*np.cos(phi)
 
# y = a*phi*np.sin(phi)
 
# dr = (np.diff(x)**2 + np.diff(y)**2)**.5 # segment lengths
 
# r = np.zeros_like(x)
 
# r[1:] = np.cumsum(dr) # integrate path
 
# r_int = np.linspace(0, r.max(), 200) # regular spaced path
 
# x_int = np.interp(r_int, r, x) # integrate path
 
# y_int = np.interp(r_int, r, y)
 
99. 给定整数n和2D数组X，从X中选择可以解释为具有n度的多项分布的绘制的行，即，仅包含整数并且总和为n的行。
 
(提示: np.logical_and.reduce, np.mod)
 
In [ ]:
 
# X = np.asarray([[1.0, 0.0, 3.0, 8.0],
 
# [2.0, 0.0, 1.0, 1.0],
 
# [1.5, 2.5, 1.0, 0.0]])
 
# n = 4
 
# M = np.logical_and.reduce(np.mod(X, 1) == 0, axis=-1)
 
# M &= (X.sum(axis=-1) == n)
 
# print (X[M])
 
100. 计算1D阵列X的平均值的自举95％置信区间（即，对替换N次的阵列的元素进行重新采样，计算每个样本的平均值，然后计算均值上的百分位数）。
 
In [ ]:
 
# X = np.random.randn(100) # random 1D array
 
# N = 1000 # number of bootstrap samples
 
# idx = np.random.randint(0, X.size, (N, X.size))
 
# means = X[idx].mean(axis=1)
 
# confint = np.percentile(means, [2.5, 97.5])
 
# print (confint)

私有云论坛

-- 家庭宽带、NAS、群晖、Python、Delphi、qt c++学习讨论

100道练习带你玩转Numpy

100道练习带你玩转Numpy