What is NumPy

NumPy is an Open Source package available for Python 2 and Python 3 that provides:

ndarray: a high-performance class to represent multidimensional arrays;
vectorized functions that apply to all elements of an array without explicit looping, including linear algebra operations, random number generators and Fourier transforms;
efficient loading/saving of arrays to disk and memory-mapped files;
interfaces to integrate C, C++ and Fortran codebases.

Introduction to `ndarray`

The np.array function is the simplest way to create an ndarray:



In [1]:

    
import numpy as np
a0 = np.array([1.1, 2.2, 3.3])

The np.array function creates instances of the ndarray class:



In [22]:

    
type(a0)









    Out[22]:





numpy.ndarray

All elements of an ndarray are of the same type, which you can inspect as the .dtype property:



In [3]:

    
a0.dtype









    Out[3]:





dtype('float64')



In [24]:

    
a1 = np.array(range(10))



In [25]:

    
a1









    Out[25]:





array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])



In [26]:

    
a1.dtype









    Out[26]:





dtype('int64')



In [27]:

    
a1 = np.arange(10)
a1









    Out[27]:





array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])

NumPy can pick a type automatically, or you can assign one explicitly:



In [34]:

    
ab = np.array([1, 2, 3], dtype=np.uint8)
ab









    Out[34]:





array([1, 2, 3], dtype=uint8)

A sequence of nested sequences with equal lenghts can be used to build a multidimensional array, like this 2D array:



In [11]:

    
a2 = np.array([[10, 20], [30, 40], [50, 60]])
a2









    Out[11]:





array([[10, 20],
       [30, 40],
       [50, 60]])



In [19]:

    
at = a2.transpose()
at









    Out[19]:





array([[10, 99, 50],
       [20, 40, 60]])

Many array operations return views which share the undelying data with the source array. That's the case of the T property and the .transpose() method:



In [20]:

    
at[0, 1] = 99



In [21]:

    
a2









    Out[21]:





array([[10, 20],
       [99, 40],
       [50, 60]])

You can also use the np.zeros, np.ones and np.identity to build arrays filled with zeros and ones:



In [41]:

    
np.ones((3, 5))









    Out[41]:





array([[ 1.,  1.,  1.,  1.,  1.],
       [ 1.,  1.,  1.,  1.,  1.],
       [ 1.,  1.,  1.,  1.,  1.]])



In [42]:

    
np.identity(4)









    Out[42]:





array([[ 1.,  0.,  0.,  0.],
       [ 0.,  1.,  0.,  0.],
       [ 0.,  0.,  1.,  0.],
       [ 0.,  0.,  0.,  1.]])

The np.random module has functions to build arrays filled with random numbers using several distributions:



In [29]:

    
a3 = np.random.random((2, 3))
a3









    Out[29]:





array([[ 0.89382349,  0.94526305,  0.14999355],
       [ 0.22736863,  0.76230151,  0.75217348]])

Many scalar operations are supported, and are vectorized:



In [30]:

    
a3 * 10









    Out[30]:





array([[ 8.93823487,  9.45263051,  1.49993552],
       [ 2.2736863 ,  7.62301513,  7.52173478]])



In [47]:

    
a3 > .5









    Out[47]:





array([[ True,  True, False],
       [False,  True,  True]], dtype=bool)

Starting with Python 3.5, the @ operator does matrix multiplication (dot product):



In [32]:

    
a2 @ a3









    Out[32]:





array([[  13.48560746,   24.69866077,   16.54340509],
       [  97.5832704 ,  124.0731026 ,   44.93630081],
       [  58.33329213,   93.00124335,   52.63008632]])

If you are using Python 3.4 or older, you must use the .dot() method for the dot product:



In [33]:

    
a2.dot(a3)









    Out[33]:





array([[  13.48560746,   24.69866077,   16.54340509],
       [  97.5832704 ,  124.0731026 ,   44.93630081],
       [  58.33329213,   93.00124335,   52.63008632]])



In [ ]:

What is NumPy

Introduction to ndarray

Introduction to `ndarray`