整活NumPy

文档官网

1. Introduction

区别	NumPy ndarray数组	Python原生Array数组
长度是否固定	固定，若改变长度，则新建一个数组	长度动态
数组元素类型是否相同	必须相同	可以不同
高级数学操作	支持度高，效率高，代码量少	支持度第，效率低，代码复杂

为了实现高效的计算，很多科学计算库都使用NumPy作为计算的载体。即使是使用Python原生数组作为输入，内部也将其转化为numpy数组。

当涉及到 ndarray 时，逐个元素的操作由预编译的C代码快速执行。

• numpy的两个特性：矢量化与广播

2. Quick Start

2.1 Basic Attribute

import numpy as np
a = np.array([
    [1,2,3],
    [4,5,6]
])
print(a.ndim) # return 2, dimensions
print(a.shape) # return (2, 3)
print(a.size) # return 6, total number of elements
print(a.dtype) # return int32
print(a.itemsize) # return 4, the size of int32 is 4 Bytes
print(a.data) # return <memory at 0x000002E2DD390E48>, the memory location of a

2.2 Array Creation

Normal creation:

import numpy as np
a = np.array([2,3,4], dtype=float) # normal defination
b = np.zeros((3, 4)) # zeros array
c = np.ones((2,3,4), dtype=np.int16) # ones array
d = np.empty((2,3)) # array without initialization

Arange function:

import numpy as np
a = np.arange(10, 30, 5)
# array between 10 <= x < 30, step = 5
# output: a = [10 15 20 25]

Linspace function:

import numpy as np
a = np.linspace(0,2,9)
# array length = 9
# output: [0.   0.25 0.5  0.75 1.   1.25 1.5  1.75 2.  ]

2.3 Print Arrays

import numpy as np
a = np.linspace(0,15,16)
a.reshape((4,4))
# output: 
# [[ 0.  1.  2.  3.]
#  [ 4.  5.  6.  7.]
#  [ 8.  9. 10. 11.]
#  [12. 13. 14. 15.]]

如果数组过大，会显示不全。通过如下指令强制显示完整。

import sys
np.set_printoptions(threshold=sys.maxsize)

2.4 Basic Operations

矢量化运算

a ** 2 # 乘方
a * b # 点积
a @ b # 向量积， 或者 a.dot(b)

*= , += 等原地运算符也存在。

还有一系列的 min, max, sum, cumsum, exp, sqrt, add等函数，就不再一一介绍了。

2.5 indexing, slicing, iterating

a[index]
b[start:end:step]

注意：若使用

b = a[1:3]

则获得的是原数组，而不是原数组的拷贝，这与原生python array不同。

2.6 reshape

以下过程均为原地变换。

a.ravel() # 转化为一维数组
a.reshape((4,4))

2.7 stacking together different arrays

a = np.floor(10*rg.random((2,2)))
b = np.floor(10*rg.random((2,2)))
b = np.vstack(a,b) # 纵向堆叠
d = np.hstack(a,b) # 水平堆叠

2.8 spliting one array into several smaller ones

np.hsplit(a, 3) # 将a水平分为三份
np.vsplit(a, 3) # 将a纵向分为三份

2.9 copy

shallow copy

import numpy as np
a = np.linspace(0,15,16)
c = a # 这个过程不是浅拷贝，它连新的np.ndarray对象都没有创建，两者的id值完全相同
c = a.view() # 浅拷贝，创建了新的np.ndarray对象，两者的id值不同，但是内部的数组引用完全相同
c = a[:] # 这个与上面的view()等效

deep copy

import numpy as np
a = np.linspace(0, 15, 16)
b = a.copy()
print(a.base is b.base) # return false, 该深拷贝对多维数组也有效

2.10 Indexing

Indexing with Arrays od Indices

import numpy as np
a = np.arange(0,15,1)
b = np.array([1,6,8])
print(a[b]) # return [1 6 8]

e.g. 获取矩阵中的纵向最大值

import numpy as np
time = np.linspace(20, 145, 5)
data = np.sin(np.arange(20)).reshape(5,4)
ind = data.argmax(axis=0)
time_max = time[ind]
print(time_max)

a[[1,3,4]] = 0 # 令这三个索引位的元素为0

Indexing with boolean values

e.g. 输出矩阵中大于4的元素，并修改为0

a = np.arrange(12).reshape(3,4)
b = a > 4 # b 是一个boolean 矩阵
print(a[b])
a[b] = 0

ix_() function

import numpy as np
a = np.arange(3)
b = np.arange(4)
c = np.arange(5)
ax, bx, cx = np.ix_(a, b, c)
# ax.size() = (3,1,1)
# bx.size() = (1,4,1)
# cx.size() = (1,1,5)

use string index

2.11 Linear Algebra

Simple Array Operations

1. transpose转置

import numpy as np
a = np.array([
    [1, 2],
    [3, 4]
])
print(a.transpose())
'''
	转置操作，return 
	[
		[1 3]
		[2 4]
	]
'''

2. 求逆

np.linalg.inv(a)
'''
	对矩阵a求逆，return 
	[
		[-2.  1. ]
		[ 1.5 0.5]
	]
'''

3. 单位矩阵

u = np.eye(2)
'''
	[
		[1 0]
		[0 1]
	]
'''

4. 内积

j @ j

5. 迹

np.trace(u) # return 2.0

6. 求解线性方程组

a = np.array([
    [1, 2],
    [3, 4]
])
y = np.array([[5.], [7.]])
np.linalg.solve(a, y)

7. 特征向量和特征值

import numpy as np
j = np.array([[0.0, -1.0], [1.0, 0.0]])
print(np.linalg.eig(j))
'''
	return (array([0.+1.j, 0.-1.j]), array([[0.70710678+0.j        , 0.70710678-0.j        ],
       		[0.        -0.70710678j, 0.        +0.70710678j]]))
'''

2.12 Tricks and Tips

Automatic Compute Reshaping Size

a = np.arange(30)
b = a.reshape((2, -1, 3))
print(b.shape) # return (2, 5, 3)

2.13 Histograms

import numpy as np
import matplotlib.pyplot as plt

rg = np.random.default_rng(1)
mu, sigma = 2, 0.5
v = rg.normal(mu,sigma,10000)
plt.hist(v, bins=50, density=1) # use plt api draw histograms
plt.show() 

(n, bins) = np.histogram(v, bins=50, density=True)
plt.plot(.5 * (bins[1:] + bins[:-1]), n) # use np api draw histograms
plt.show()