创建ndarray
In [1]: data = [[1,2,3,4],[5,6,7,8]]
In [2]: arry = np.array(data)
In [3]: arry
Out[3]:
array([[1, 2, 3, 4],
[5, 6, 7, 8]])
查看ndarray的维度
In [4]: arry.ndim
Out[4]: 2
输出ndarray的行数和列数
In [5]: arry.shape
Out[5]: (2, 4)
zeros和ones可以创建指定长度或形状的全是零和一的数组
In [15]: np.zeros((3,4))
Out[15]:
array([[0., 0., 0., 0.],
[0., 0., 0., 0.],
[0., 0., 0., 0.]])
empty会产生一些未初始化的垃圾值
In [17]: np.empty((2,3,2))
Out[17]:
array([[[0., 0.],
[0., 0.],
[0., 0.]],
[[0., 0.],
[0., 0.],
[0., 0.]]])
数组和标量之间的运算
In [1]: arry = np.array([[1,2,3,4],[5,6,7,8]])
In [2]: arry * arry
Out[2]:
array([[ 1, 4, 9, 16],
[25, 36, 49, 64]])
In [3]: arry - arry
Out[3]:
array([[0, 0, 0, 0],
[0, 0, 0, 0]])
In [4]: 1 / arry
Out[4]:
array([[1, 0, 0, 0],
[0, 0, 0, 0]])
In [5]: arry ** 0.5
Out[5]:
array([[1. , 1.41421356, 1.73205081, 2. ],
[2.23606798, 2.44948974, 2.64575131, 2.82842712]])
基本索引和切片
In [6]: arr = np.arange(15)
In [7]: arr
Out[7]: array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14])
In [8]: arr[5]
Out[8]: 5
In [9]: arr[5:8]
Out[9]: array([5, 6, 7])
In [10]: arr[5:8] = 12
In [11]: arr
Out[11]: array([ 0, 1, 2, 3, 4, 12, 12, 12, 8, 9, 10, 11, 12, 13, 14])
In [12]: arr_slice = arr[5:8]
In [13]: arr_slice[1] = 1234
In [14]: arr
Out[14]:
array([ 0, 1, 2, 3, 4, 12, 1234, 12, 8, 9, 10,
11, 12, 13, 14])
In [19]: arr2d = np.array([[1,2,3],[4,5,6],[7,8,9]])
> 将一个标量赋值给一个切片时，该值将自动传播到选中的区域。跟列表最重要的区别在于，数组切片是原始数组的视图，这意味着数据不会被复制，视图上的修改会直接反应到原始数组。主要原因如果numpy坚持将数据复制来复制去会产生性能和内存问题
In [20]: arr2d
Out[20]:
array([[1, 2, 3],
[4, 5, 6],
[7, 8, 9]])
In [21]: arr2d[:2]
Out[21]:
array([[1, 2, 3],
[4, 5, 6]])
In [22]: arr2d[:2,:1]
Out[22]:
array([[1],
[4]])
In [23]: arr2d[:2,1:]
Out[23]:
array([[2, 3],
[5, 6]])
In [24]: arr2d[:,:1]
Out[24]:
array([[1],
[4],
[7]])
>可以一次性传入多个切片;只有冒号时，表示选取整个轴
布尔型索引
In [29]: names
Out[29]: array(['Bob', 'Joe', 'Will', 'Bob', 'Will', 'Joe', 'Joe'], dtype='|S4')
In [33]: data = np.random.randn(7,4)
In [34]: data
Out[34]:
array([[ 1.16675978, 1.92107701, 0.26941017, -1.19809428],
[ 0.07649339, -0.91997008, -0.21832289, 1.07434831],
[-1.98754599, -0.60850528, 0.06302264, -0.63377773],
[ 0.32234302, 0.32310855, 0.73595001, 0.24015727],
[-1.09218583, -0.08907211, -1.51447014, 0.53083635],
[-0.27048015, -0.09299773, 0.75937323, -0.71705501],
[ 1.23737272, -0.08242024, -2.3591149 , 0.64462961]])
In [37]: data[names=='Bob']
Out[37]:
array([[ 1.16675978, 1.92107701, 0.26941017, -1.19809428],
[ 0.32234302, 0.32310855, 0.73595001, 0.24015727]])
>布尔型数组的长度必须和被索引轴相同
In [38]: data[names=='Bob',2:]
Out[38]:
array([[ 0.26941017, -1.19809428],
[ 0.73595001, 0.24015727]])
In [40]: data[names != 'Bob']
Out[40]:
array([[ 0.07649339, -0.91997008, -0.21832289, 1.07434831],
[-1.98754599, -0.60850528, 0.06302264, -0.63377773],
[-1.09218583, -0.08907211, -1.51447014, 0.53083635],
[-0.27048015, -0.09299773, 0.75937323, -0.71705501],
[ 1.23737272, -0.08242024, -2.3591149 , 0.64462961]])
In [41]: mask = (names == 'Bob') | (names == 'Will')
In [42]: data[mask]
Out[42]:
array([[ 1.16675978, 1.92107701, 0.26941017, -1.19809428],
[-1.98754599, -0.60850528, 0.06302264, -0.63377773],
[ 0.32234302, 0.32310855, 0.73595001, 0.24015727],
[-1.09218583, -0.08907211, -1.51447014, 0.53083635]])
>支持与或非的布尔运算
In [43]: data[data<0] = 0
In [44]: data
Out[44]:
array([[1.16675978, 1.92107701, 0.26941017, 0. ],
[0.07649339, 0. , 0. , 1.07434831],
[0. , 0. , 0.06302264, 0. ],
[0.32234302, 0.32310855, 0.73595001, 0.24015727],
[0. , 0. , 0. , 0.53083635],
[0. , 0. , 0.75937323, 0. ],
[1.23737272, 0. , 0. , 0.64462961]])
In [45]:
>把小于零的值制成零
花式索引
In [45]: arr = np.empty((8,4))
In [48]: for i in range(8):
...: arr[i] = i
...:
In [49]: arr
Out[49]:
array([[0., 0., 0., 0.],
[1., 1., 1., 1.],
[2., 2., 2., 2.],
[3., 3., 3., 3.],
[4., 4., 4., 4.],
[5., 5., 5., 5.],
[6., 6., 6., 6.],
[7., 7., 7., 7.]])
In [50]: arr[[4,3,0,6]]
Out[50]:
array([[4., 4., 4., 4.],
[3., 3., 3., 3.],
[0., 0., 0., 0.],
[6., 6., 6., 6.]])
In [51]: arr[[-3,-5,-7]]
Out[51]:
array([[5., 5., 5., 5.],
[3., 3., 3., 3.],
[1., 1., 1., 1.]])
>负数索引是从末尾开始选取
In [52]: arr = np.arange(32).reshape((8,4))
In [53]: arr
Out[53]:
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, 30, 31]])
In [54]: arr[[1,5,7,2],[0,3,1,2]]
Out[54]: array([ 4, 23, 29, 10])
>这种结果不是我们所需要的
In [55]: arr[[1,5,7,2]][:,[0,3,1,2]]
Out[55]:
array([[ 4, 7, 5, 6],
[20, 23, 21, 22],
[28, 31, 29, 30],
[ 8, 11, 9, 10]]
In [57]: arr[np.ix_([1,5,7,2],[0,3,1,2])]
Out[57]:
array([[ 4, 7, 5, 6],
[20, 23, 21, 22],
[28, 31, 29, 30],
[ 8, 11, 9, 10]])
>np.ix_函数可以把一位函数变成索引选择器
数组转置和轴对换
In [60]: arr
Out[60]:
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14]])
In [61]: arr.T
Out[61]:
array([[ 0, 5, 10],
[ 1, 6, 11],
[ 2, 7, 12],
[ 3, 8, 13],
[ 4, 9, 14]])
相关概念
1. 矢量化:即使不用编写循环即可对数据进行计算.
2.不同大小数组之间的运算叫广播
3.花式索引指利用整数数组进行索引;花式索引跟切片不一样，它总是将数据复制到新数组.