Numpy stay python Is a very common bag , Whether it's machine learning collocation pandas, Or data visualization collocation pylab It's a very normal match .

Numpy

numpy Official Chinese document of ：NumPy chinese

NumPy It's using Python Basic software package for Scientific Computing . Among other things , It includes ：

• Powerful N Dimensional array object ;
• Precise broadcast function ;
• Integrate C/C+ and Fortran Code tools ;
• Powerful Linear Algebra 、 Fourier transform and random number function .

More simply ,Numpy yes Python Of Matlab Math package . Use it ,python The matrix vector can be calculated more simply and conveniently .

Generally speaking , We refer to the package and abbreviate it to np：

import numpy as np

Numpy Of ndarry object

ndarray The creation of

Numpy The most important data type in is ：N Dimensional array object ndarray. It's a collection of data of the same type , With 0 The subscript is to start indexing the elements in the collection .

It has the following two characteristics ：

• ndarray Object is a multidimensional array used to hold elements of the same type ;
• ndarray Each element in has an area of the same storage size in memory .

Create a ndarray object ：

np.array(object, dtype = None, copy = True, order = None, subok = False, ndmin = 0)

Parameter description ：

name	describe
object	An array or nested sequence of numbers
dtype	Data type of array element , Optional
copy	Whether the object needs to be copied , Optional
order	Create an array style ,C In the direction of the line ,F For column direction ,A In any direction ( Default )
subok	By default, it returns an array consistent with the base class type
ndmin	Specifies the minimum dimension of the generated array

generally speaking , You don't need to remember so many optional parameters ：

import numpy as np
if __name__ == "__main__":
x = np.array([1, 2, 3])
print(x)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[1 2 3]

Of course , In addition to the previous array Method creation ndarray Beyond the object , There are several other ways to create ：

np.empty(shape, dtype = float, order = 'C') # Uninitialized array of specified shapes 
np.zeros(shape, dtype = float, order = 'C') # All of the specified shapes 0 Array 
np.ones(shape, dtype = None, order = 'C') # All of the specified shapes 1 Array 
np.arange(start = 0, stop, step = 1, dtype) # From the starting value to the ending value ( It doesn't contain ) when , Create an array from a range in steps 
np.linspace(start, stop, num = 50, endpoint = True, retstep = False, dtype = None) # From the starting value to the ending value ( Default includes ) Create a one-dimensional array of isometric arrays 
np.logspace(start, stop, num = 50, endpoint = True, base = 10.0, dtype = None) # From the starting value to the ending value ( Default includes ) Create a one-dimensional array of proportional arrays 

Numpy Also provided from python Other types are converted directly to ndarray The way ：

np.asarray(a, dtype = None, order = None) # Tabular form 
np.frombuffer(buffer, dtype = float, count = -1, offset = 0) # Read in as a stream 
np.fromiter(iterable, dtype, count=-1) # From iteratable objects , Read in as an iterator 

for example ：

import numpy as np
if __name__ == "__main__":
a = [[1, 2 ,3], [4, 5]]
b = 'Hello World'
c = iter(range(5))
x = np.asarray(a)
y = np.frombuffer(b, dtype = 'S1')
z = np.fromiter(c, dtype = float)
print(x)
print(y)
print(z)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[list([1, 2, 3]) list([4, 5])]
['H' 'e' 'l' 'l' 'o' ' ' 'W' 'o' 'r' 'l' 'd']
[0. 1. 2. 3. 4.]

ndarray Data structure of

Numpy Support a lot of data types , Here's a simple list ：

name	describe
bool_	Boolean data type (True perhaps False)
int_/int8/int16/int32/int64	Signed integers
uint8/uint16/uint32/uint64	Unsigned integer
float_/float16/float32/float64	Floating point numbers
complex_/complex64/complex128	The plural

But if it's a custom data type , It needs to pass dtype To make sure ：

numpy.dtype(object, align, copy)

Parameter description ：

name	describe
object	Data type object to convert to
align	If true, Fill in the fields to make them look like C The structure of the body
copy	Copy dtype object , If false, Is a reference to a built-in data type object

for example , You can create a student The object of ：

import numpy as np
if __name__ == "__main__":
student = np.dtype([('name', 'S20'), ('age', 'i8'), ('score', 'f4')])
a = np.array([('zhangsan', 18, 80), ('lisi', 19, 85)], dtype=student)
print(a)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[('zhangsan', 18, 80.) ('lisi', 19, 85.)]

ndarray Properties of

ndarray There are two very common properties ,shape and size.shape Represents the dimensions of an array , For two-dimensional arrays , It's the number of rows and columns ;size Represents the total number of array elements , For two-dimensional arrays , It's the multiplication of the number of rows and columns .

for example ：

import numpy as np
if __name__ == "__main__":
a = np.array([[1, 2, 3], [4, 5, 6]])
print(a.shape)
print(a.size)

Run the script ：

[email protected]:~/test$ python numpy_test.py
(2, 3)
6

Of course ,ndarray Objects provide two ways to do this without changing the content of the data , Change the format of an array . But the two ways are different ：

import numpy as np
if __name__ == "__main__":
a = np.array([[1, 2, 3], [4, 5, 6]])
a.shape = (3, 2) # Directly change the noumenon 
print(a)
b = a.reshape(2, 3) # The noumenon does not change , Return the changed object to 
print(b)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[[1 2]
[3 4]
[5 6]]
[[1 2 3]
[4 5 6]]

ndarray Content access for

ndarray The contents of an object can be accessed and modified by index or slice , And python in list The slice operation is the same .

ndarray It can be sliced based on the subscript , It's fine too Through the built-in slice function , And set up start,stop And step Parameters , Cut a new array from the original array .

for example ：

import numpy as np
if __name__ == "__main__":
a = np.arange(10)
b = a[1:7:1]
s = slice(1,7,1)
c = a[s]
print(a)
print(b)
print(c)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[0 1 2 3 4 5 6 7 8 9]
[1 2 3 4 5 6]
[1 2 3 4 5 6]

For colons : The explanation of ：

• If it's a one-dimensional array , If you put only one parameter , Such as [2], A single element corresponding to the index... Will be returned . If [2:], Indicates that all items from the beginning of the index will be extracted . If two parameters are used , Such as [2:7], Then extract two indexes ( Does not include stop indexing ) Between the items ;
• If it's a multidimensional array , Use , Distinguishing dimensions .

for example ：

import numpy as np
if __name__ == "__main__":
a = np.arange(25)
a.shape = (5, 5)
b = a[1:4, 2:4]
print(a)
print(b)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[[ 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]]
[[ 7 8]
[12 13]
[17 18]]

ndarray Except for slicing based on subscripts , There are also some advanced indexing methods , Like Boolean indexes 、 Fancy index .

for example ：

import numpy as np
if __name__ == "__main__":
a = np.arange(25)
a.shape = (5, 5)
b = a[a > 6]
c = a[[3, 2, 4]]
print(a)
print(b)
print(c)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[[ 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]]
[ 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24]
[[15 16 17 18 19]
[10 11 12 13 14]
[20 21 22 23 24]]

other

Judge whether the element objects are all NaN：

np.isnan(...)

take ndarray The object of the is changed to list：

obj.tolist()

Numpy Broadcast of

If two ndarray：a and b The same shape , The meet a.shape==b.shape, that a And b The result is a And b Arrays do arithmetic operations on bits . This requires The dimensions are the same , And the length of each dimension is the same .

for example ：

import numpy as np
if __name__ == "__main__":
a = np.array([1, 2, 3])
b = np.array([1, 2, 3])
c = a + b
d = a * b
print(c)
print(d)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[2 4 6]
[1 4 9]

and , Radio is Numpy For different shapes (shape) Of ndarray How to do numerical calculation , Yes ndarray To perform arithmetic operations on the corresponding elements .

How is it a corresponding element ？

although , Broadcasting is about different shapes (shape) for , But in fact, two conditions have to be met ： Same number of columns , There is a line number of 1. Under this premise , The elements of the same column in each row are the corresponding elements .

It's impossible to read the text directly , Take a look at an example ：

import numpy as np
if __name__ == "__main__":
a = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
b = np.array([1, 2, 3])
c = a + b
d = a * b
print(c)
print(d)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[[ 2 4 6]
[ 5 7 9]
[ 8 10 12]]
[[ 1 4 9]
[ 4 10 18]
[ 7 16 27]]

The so-called broadcasting is ： When the number of columns is the same , The number of rows is 1 Of ndarray Will expand the line operation , The content of the increased number of lines is the same as that of the original line .

Extension can be realized by tile Function implementation ：

np.tile(obj, ( That's ok , Column )) # Repeat on the row and column a certain number of times 

therefore , The above broadcast can also be replaced by the following ways ：

import numpy as np
if __name__ == "__main__":
a = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
b = np.array([1, 2, 3])
bb = np.tile(b, (3, 1))
c = a + bb
d = a * bb
print(bb)
print(c)
print(d)

Run the script ：

[email protected]:~/test$ python numpy_test.py
[[1 2 3]
[1 2 3]
[1 2 3]]
[[ 2 4 6]
[ 5 7 9]
[ 8 10 12]]
[[ 1 4 9]
[ 4 10 18]
[ 7 16 27]]

ndarray Function of

ndarray Provides a lot of mathematical functions 、 Arithmetic functions 、 Sorting function , In order to do the calculation .

ndarray The mathematical function of , for example ：

np.pi # PI 
np.sin(obj) # Trigonometry 
np.cos(obj)
np.tan(obj)
np.arcsin(obj) # Inverse trigonometry 
np.arccos(obj)
np.arctan(obj)
np.degrees(obj) # Convert radian value to angle value 
np.around(obj, decimals) # return ndarray The rounding value of each element ,decimals Is the rounding number of decimal places , The default is 0
np.floor(obj) # Rounding down 
np.ceil(obj) # Rounding up 

ndarray The arithmetic function of , for example ：

np.add(obj1, obj2) # Addition, subtraction, multiplication and division , And +-*/ The effect is consistent , Need to comply with broadcasting principles 
np.subtract(obj1, obj2)
np.multiply(obj1, obj2)
np.divide(obj1, obj2)
np.mod(obj1, obj2) # Remainder operation 
np.reciprocal(obj) # Take the reciprocal of the elements 
np.power(obj1, obj2) # Before calculation, the parameter is the base , The latter parameter is a power value 

ndarray The sort function of , for example ：

np.sort(obj, axis=1, kind='quicksort', order)

Parameter description ：

name	describe
obj	An array or nested sequence of numbers
axis	axis=0 Sort by column ,axis=1 Sort by row
kind	‘quicksort’、‘mergesort’、‘heapsort’
order	If the array contains fields , The fields to be sorted

Related reading

python Medium numpy modular