基础
numpy的主要目标是同类型的多维数组。是一个元素表,数据类型都一样,用一个非负整数元组来获取数据。numpy 纬度都叫做axes(坐标轴)
比如,一个三维空间的坐标点[1, 2, 1]
,只有一个轴,其实就是说一维的。
numpy 数组类型都叫做ndarry,别名叫做array,这个array和python的array.array非常不一样,因为numpy的数组有更多的功能。
- 数组纬度
ndarray.ndim
- 数组形状
ndarray.shape
- 数组所有元素个数
ndarray.size
- 数组数据类型
ndarray.dtype
- 数组字节大小
ndarray.itemsize
创建数组
有好几种方法创建数组
- 从python列表创建数组
a = np.array([2,3,4])
c = np.array( [ [1,2], [3,4] ], dtype=complex )
指定数据类型np.zeros( (3,4) )
创建都是0的数组,默认类型float64np.ones( (2,3,4), dtype=np.int16 )
创建都是1的数组,指定数据类型np.empty( (2,3) )
创建空数组,数据数据是随机的,具体数据看内存情况np.arange( 0, 2, 0.3 )
创建序列np.linspace( 0, 2, 9 ) array([ 0. , 0.25, 0.5 , 0.75, 1. , 1.25, 1.5 , 1.75, 2. ])
创建某段之间,指定多少个数
打印数组
当你打印数组,numpy展示它的结果是嵌套列表,
>>> a = np.arange(6) # 1d array
>>> print(a)
[0 1 2 3 4 5]
>>>
>>> b = np.arange(12).reshape(4,3) # 2d array
>>> print(b)
[[ 0 1 2]
[ 3 4 5]
[ 6 7 8]
[ 9 10 11]]
>>>
>>> c = np.arange(24).reshape(2,3,4) # 3d array
>>> print(c)
[[[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]]
[[12 13 14 15]
[16 17 18 19]
[20 21 22 23]]]
See below to get more details on reshape.
打印规律是,最后一维数据是从左到右,剩下的都是从上往下,
基础操作
>>> a = np.array( [20,30,40,50] )
>>> b = np.arange( 4 )
>>> b
array([0, 1, 2, 3])
>>> c = a-b
>>> c
array([20, 29, 38, 47])
>>> b**2
array([0, 1, 4, 9])
>>> 10*np.sin(a)
array([ 9.12945251, -9.88031624, 7.4511316 , -2.62374854])
>>> a<35
array([ True, True, False, False])
numpy数组计算是广播形式的
>>> A = np.array( [[1,1],
... [0,1]] )
>>> B = np.array( [[2,0],
... [3,4]] )
>>> A * B # elementwise product
array([[2, 0],
[0, 4]])
>>> A @ B # matrix product
array([[5, 4],
[3, 4]])
>>> A.dot(B) # another matrix product
array([[5, 4],
[3, 4]])
矩阵点乘,矩阵乘积
>>> a = np.ones((2,3), dtype=int)
>>> b = np.random.random((2,3))
>>> a *= 3
>>> a
array([[3, 3, 3],
[3, 3, 3]])
>>> b += a
>>> b
array([[ 3.417022 , 3.72032449, 3.00011437],
[ 3.30233257, 3.14675589, 3.09233859]])
>>> a += b # b is not automatically converted to integer type
Traceback (most recent call last):
...
TypeError: Cannot cast ufunc add output from dtype('float64') to dtype('int64') with casting rule 'same_kind'
比如*=,+=这些操作符就是在现有数组上进行修改,而不是创建一个新的,所以其实省内存了
>>> a = np.ones(3, dtype=np.int32)
>>> b = np.linspace(0,pi,3)
>>> b.dtype.name
'float64'
>>> c = a+b
>>> c
array([ 1. , 2.57079633, 4.14159265])
>>> c.dtype.name
'float64'
>>> d = np.exp(c*1j)
>>> d
array([ 0.54030231+0.84147098j, -0.84147098+0.54030231j,
-0.54030231-0.84147098j])
>>> d.dtype.name
'complex128'
numpy计算精度向精度高的一方看齐
>>> a = np.random.random((2,3))
>>> a
array([[ 0.18626021, 0.34556073, 0.39676747],
[ 0.53881673, 0.41919451, 0.6852195 ]])
>>> a.sum()
2.5718191614547998
>>> a.min()
0.1862602113776709
>>> a.max()
0.6852195003967595
一元操作,这个是针对所有数据的
>>> b = np.arange(12).reshape(3,4)
>>> b
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]])
>>>
>>> b.sum(axis=0) # sum of each column
array([12, 15, 18, 21])
>>>
>>> b.min(axis=1) # min of each row
array([0, 4, 8])
>>>
>>> b.cumsum(axis=1) # cumulative sum along each row
array([[ 0, 1, 3, 6],
[ 4, 9, 15, 22],
[ 8, 17, 27, 38]])
通过指定纬度,还能够进行一元操作
通用函数
>>> B = np.arange(3)
>>> B
array([0, 1, 2])
>>> np.exp(B)
array([ 1. , 2.71828183, 7.3890561 ])
>>> np.sqrt(B)
array([ 0. , 1. , 1.41421356])
>>> C = np.array([2., -1., 4.])
>>> np.add(B, C)
array([ 2., 0., 6.])
比如
all, any, apply_along_axis, argmax, argmin, argsort, average, bincount, ceil, clip, conj, corrcoef, cov, cross, cumprod, cumsum, diff, dot, floor, inner, inv, lexsort, max, maximum, mean, median, min, minimum, nonzero, outer, prod, re, round, sort, std, sum, trace, transpose, var, vdot, vectorize, where
索引,切片,迭代
以为数组可以被索引,切片,迭代,基本上和python的列表一致
>>> a = np.arange(10)**3
>>> a
array([ 0, 1, 8, 27, 64, 125, 216, 343, 512, 729])
>>> a[2]
8
>>> a[2:5]
array([ 8, 27, 64])
>>> a[:6:2] = -1000 # equivalent to a[0:6:2] = -1000; from start to position 6, exclusive, set every 2nd element to -1000
>>> a
array([-1000, 1, -1000, 27, -1000, 125, 216, 343, 512, 729])
>>> a[ : :-1] # reversed a
array([ 729, 512, 343, 216, 125, -1000, 27, -1000, 1, -1000])
>>> for i in a:
... print(i**(1/3.))
...
nan
1.0
nan
3.0
nan
5.0
6.0
7.0
8.0
9.0
多维数组,每一个纬度都有一个索引,这些索引就是被逗号隔开的元组
>>> def f(x,y):
... return 10*x+y
...
>>> b = np.fromfunction(f,(5,4),dtype=int)
>>> b
array([[ 0, 1, 2, 3],
[10, 11, 12, 13],
[20, 21, 22, 23],
[30, 31, 32, 33],
[40, 41, 42, 43]])
>>> b[2,3]
23
>>> b[0:5, 1] # each row in the second column of b
array([ 1, 11, 21, 31, 41])
>>> b[ : ,1] # equivalent to the previous example
array([ 1, 11, 21, 31, 41])
>>> b[1:3, : ] # each column in the second and third row of b
array([[10, 11, 12, 13],
[20, 21, 22, 23]])
当更少索引提供的时候,缺失的索引被认为是针对整体的
>>> b[-1] # the last row. Equivalent to b[-1,:]
array([40, 41, 42, 43])
dots (…) 代表许多冒号(colons),
- x[1,2,…] to x[1,2,:,:,:],
- x[…,3] to x[:,:,:,:,3]
- x[4,…,5,:] to x[4,:,:,5,:].
>>> c = np.array( [[[ 0, 1, 2], # a 3D array (two stacked 2D arrays)
... [ 10, 12, 13]],
... [[100,101,102],
... [110,112,113]]])
>>> c.shape
(2, 2, 3)
>>> c[1,...] # same as c[1,:,:] or c[1]
array([[100, 101, 102],
[110, 112, 113]])
>>> c[...,2] # same as c[:,:,2]
array([[ 2, 13],
[102, 113]])
迭代就是针对多维数组的第一维来遍历的
>>> for row in b:
... print(row)
...
[0 1 2 3]
[10 11 12 13]
[20 21 22 23]
[30 31 32 33]
[40 41 42 43]
如果想遍历所有数据,
>>> for element in b.flat:
... print(element)
...
0
1
2
3
10
11
12
13
20
21
22
23
30
31
32
33
40
41
42
43
numpy.newaxis一个方便的别名None
>>> newaxis is None
True
>>> x = np.arange(3)
>>> x
array([0, 1, 2])
>>> x[:, newaxis]
array([[0],
[1],
[2]])
>>> x[:, newaxis, newaxis]
array([[[0]],
[[1]],
[[2]]])
>>> x[:, newaxis] * x
array([[0, 0, 0],
[0, 1, 2],
[0, 2, 4]])
形状操作
>>> a.ravel() # returns the array, flattened
array([ 2., 8., 0., 6., 4., 5., 1., 1., 8., 9., 3., 6.])
>>> a.reshape(6,2) # returns the array with a modified shape
array([[ 2., 8.],
[ 0., 6.],
[ 4., 5.],
[ 1., 1.],
[ 8., 9.],
[ 3., 6.]])
>>> a.T # returns the array, transposed
array([[ 2., 4., 8.],
[ 8., 5., 9.],
[ 0., 1., 3.],
[ 6., 1., 6.]])
>>> a.T.shape
(4, 3)
>>> a.shape
(3, 4)
这些函数并不改变原始数据形状
多个数组可以从不同维进行叠加
>>> a = np.floor(10*np.random.random((2,2)))
>>> a
array([[ 8., 8.],
[ 0., 0.]])
>>> b = np.floor(10*np.random.random((2,2)))
>>> b
array([[ 1., 8.],
[ 0., 4.]])
>>> np.vstack((a,b))
array([[ 8., 8.],
[ 0., 0.],
[ 1., 8.],
[ 0., 4.]])
>>> np.hstack((a,b))
array([[ 8., 8., 1., 8.],
[ 0., 0., 0., 4.]])
horizontal水平的
vertical垂直的
>>> from numpy import newaxis
>>> np.column_stack((a,b)) # with 2D arrays
array([[ 8., 8., 1., 8.],
[ 0., 0., 0., 4.]])
>>> a = np.array([4.,2.])
>>> b = np.array([3.,8.])
>>> np.column_stack((a,b)) # returns a 2D array
array([[ 4., 3.],
[ 2., 8.]])
>>> np.hstack((a,b)) # the result is different
array([ 4., 2., 3., 8.])
>>> a[:,newaxis] # this allows to have a 2D columns vector
array([[ 4.],
[ 2.]])
>>> np.column_stack((a[:,newaxis],b[:,newaxis]))
array([[ 4., 3.],
[ 2., 8.]])
>>> np.hstack((a[:,newaxis],b[:,newaxis])) # the result is the same
array([[ 4., 3.],
[ 2., 8.]])
复制和视图
当我操作一个数组的时候,数据有时候被拷贝到一个新的数组,有时候不会,这是一个混淆源对于刚开始入门的人。
没有拷贝
>>> a = np.arange(12)
>>> b = a # no new object is created
>>> b is a # a and b are two names for the same ndarray object
True
>>> b.shape = 3,4 # changes the shape of a
>>> a.shape
(3, 4)
python传递可变对象,函数调用也没有拷贝
>>> def f(x):
... print(id(x))
...
>>> id(a) # id is a unique identifier of an object
148293216
>>> f(a)
148293216
浅拷贝
不同的数组对象可以共享相同数据,view方法创建了一个新的数组对象,看起来像是相同数据
>>> c = a.view()
>>> c is a
False
>>> c.base is a # c is a view of the data owned by a
True
>>> c.flags.owndata
False
>>>
>>> c.shape = 2,6 # a's shape doesn't change
>>> a.shape
(3, 4)
>>> c[0,4] = 1234 # a's data changes
>>> a
array([[ 0, 1, 2, 3],
[1234, 5, 6, 7],
[ 8, 9, 10, 11]])
对一个数组切片返回它的浅拷贝
>>> s = a[ : , 1:3] # spaces added for clarity; could also be written "s = a[:,1:3]"
>>> s[:] = 10 # s[:] is a view of s. Note the difference between s=10 and s[:]=10
>>> a
array([[ 0, 10, 10, 3],
[1234, 10, 10, 7],
[ 8, 10, 10, 11]])
深拷贝
>>> d = a.copy() # a new array object with new data is created
>>> d is a
False
>>> d.base is a # d doesn't share anything with a
False
>>> d[0,0] = 9999
>>> a
array([[ 0, 10, 10, 3],
[1234, 10, 10, 7],
[ 8, 10, 10, 11]])
函数和方法概览
Here is a list of some useful NumPy functions and methods names ordered in categories. See Routines for the full list.
创建列表
arange, array, copy, empty, empty_like, eye, fromfile, fromfunction, identity, linspace, logspace, mgrid, ogrid, ones, ones_like, r, zeros, zeros_like
转换方法
ndarray.astype, atleast_1d, atleast_2d, atleast_3d, mat
操作
array_split, column_stack, concatenate, diagonal, dsplit, dstack, hsplit, hstack, ndarray.item, newaxis, ravel, repeat, reshape, resize, squeeze, swapaxes, take, transpose, vsplit, vstack
查询
all, any, nonzero, where
排序
argmax, argmin, argsort, max, min, ptp, searchsorted, sort
操作
choose, compress, cumprod, cumsum, inner, ndarray.fill, imag, prod, put, putmask, real, sum
基本统计
cov, mean, std, var
基本线性代数
cross, dot, outer, linalg.svd, vdot
深入学习
广播规则
数组索引
>>> a = np.arange(12)**2 # the first 12 square numbers
>>> i = np.array( [ 1,1,3,8,5 ] ) # an array of indices
>>> a[i] # the elements of a at the positions i
array([ 1, 1, 9, 64, 25])
>>>
>>> j = np.array( [ [ 3, 4], [ 9, 7 ] ] ) # a bidimensional array of indices
>>> a[j] # the same shape as j
array([[ 9, 16],
[81, 49]])
布尔数组索引
>>> a = np.arange(12).reshape(3,4)
>>> b = a > 4
>>> b # b is a boolean with a's shape
array([[False, False, False, False],
[False, True, True, True],
[ True, True, True, True]])
>>> a[b] # 1d array with the selected elements
array([ 5, 6, 7, 8, 9, 10, 11])
这个性质特别好用在赋值上
>>> a[b] = 0 # All elements of 'a' higher than 4 become 0
>>> a
array([[0, 1, 2, 3],
[4, 0, 0, 0],
[0, 0, 0, 0]])
线性代数
简单的数组操作
>>> import numpy as np
>>> a = np.array([[1.0, 2.0], [3.0, 4.0]])
>>> print(a)
[[ 1. 2.]
[ 3. 4.]]
>>> a.transpose()
array([[ 1., 3.],
[ 2., 4.]])
>>> np.linalg.inv(a)
array([[-2. , 1. ],
[ 1.5, -0.5]])
>>> u = np.eye(2) # unit 2x2 matrix; "eye" represents "I"
>>> u
array([[ 1., 0.],
[ 0., 1.]])
>>> j = np.array([[0.0, -1.0], [1.0, 0.0]])
>>> j @ j # matrix product
array([[-1., 0.],
[ 0., -1.]])
>>> np.trace(u) # trace
2.0
>>> y = np.array([[5.], [7.]])
>>> np.linalg.solve(a, y)
array([[-3.],
[ 4.]])
>>> np.linalg.eig(j)
(array([ 0.+1.j, 0.-1.j]), array([[ 0.70710678+0.j , 0.70710678-0.j ],
[ 0.00000000-0.70710678j, 0.00000000+0.70710678j]]))
技巧建议
为了改变数组纬度,可以忽略某一个纬度大小,这个大小可以自动被推断出来
>>> a = np.arange(30)
>>> a.shape = 2,-1,3 # -1 means "whatever is needed"
>>> a.shape
(2, 5, 3)
>>> a
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],
[24, 25, 26],
[27, 28, 29]]])