Numpy - Array creation, inspection and manipulation¶

NumPy is the primary matrix laboratory for Python. Many other libraries such as pandas, tensorflow, scikit-learn etc are built on top of this. Some tutorials to go with this cheat sheet: Numpy quick start - scipy, numpy python course eu, datacamp numpy wiki, numpy.org

In [1]:

import numpy

import numpy

Creating numpy arrays from lists¶

In [2]:

l1 = [1,2,3]
arr1 = numpy.array(l1)
arr1

l1 = [1,2,3] arr1 = numpy.array(l1) arr1

Out[2]:

array([1, 2, 3])

In [3]:

type(arr1)

type(arr1)

Out[3]:

numpy.ndarray

2D arrays¶

Create using list of lists

In [4]:

l2d = [[1,2,3],[4,5,6]]
arr2d = numpy.array(l2d)
arr2d

l2d = [[1,2,3],[4,5,6]] arr2d = numpy.array(l2d) arr2d

Out[4]:

array([[1, 2, 3],
       [4, 5, 6]])

In [5]:

type(arr2d)

type(arr2d)

Out[5]:

numpy.ndarray

3D arrays¶

Same way using list of lists

In [6]:

l3d = [[[1,2],[3,4]], [[1,2],[3,4]]]
arr3d = numpy.array(l3d)
arr3d

l3d = [[[1,2],[3,4]], [[1,2],[3,4]]] arr3d = numpy.array(l3d) arr3d

Out[6]:

array([[[1, 2],
        [3, 4]],

       [[1, 2],
        [3, 4]]])

Creating using MATLAB style syntax¶

In [10]:

mat = numpy.mat('1,2,3;4,5,6;7,8,9')
type(mat)

mat = numpy.mat('1,2,3;4,5,6;7,8,9') type(mat)

Out[10]:

numpy.matrix

In [11]:

mat

mat

Out[11]:

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

In [12]:

numpy.array(mat)

numpy.array(mat)

Out[12]:

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

Special numpy arrays¶

ones zeros and eye¶

Creating a matrix of ones

In [7]:

numpy.ones(shape=(4,4)) #pass shapre as a tuple

numpy.ones(shape=(4,4)) #pass shapre as a tuple

Out[7]:

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

zeros¶

sometimes it is useful to just get a matrix of zeros

In [8]:

numpy.zeros(shape = (2,2))

numpy.zeros(shape = (2,2))

Out[8]:

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

eye¶

eye for identity matrix - has values only in the diagonal. Identity matrices are square, 2D

In [11]:

numpy.eye(3)

numpy.eye(3)

Out[11]:

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

arange and linspace¶

arange is array range. Works same as range function in Python, but returns an array.

In [12]:

numpy.arange(start=0, stop=10, step=1) #stop is non inclusive

numpy.arange(start=0, stop=10, step=1) #stop is non inclusive

Out[12]:

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

In [13]:

numpy.arange(21) #the stop arg is the only compulsory arg

numpy.arange(21) #the stop arg is the only compulsory arg

Out[13]:

array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20])

linspace is similar, returns contiguous numbers in linearly spaced intervals. These numbers conform to uniform distribution

In [15]:

numpy.linspace(start=3, stop=21, num=7) #return 7 numbers between 3 and 21 (inclusive)

numpy.linspace(start=3, stop=21, num=7) #return 7 numbers between 3 and 21 (inclusive)

Out[15]:

array([  3.,   6.,   9.,  12.,  15.,  18.,  21.])

Random numbers¶

generate random numbers using random module

rand¶

returns numbers in normal distribution between -1 and 1

In [14]:

numpy.random.rand(3, 3) #specify shape in individual arguments

numpy.random.rand(3, 3) #specify shape in individual arguments

Out[14]:

array([[0.13036297, 0.44737951, 0.85299833],
       [0.94987992, 0.63348932, 0.42257387],
       [0.54558662, 0.17654393, 0.84926165]])

In [15]:

numpy.random.rand(3,3,3)

numpy.random.rand(3,3,3)

Out[15]:

array([[[0.30709651, 0.26703385, 0.56179506],
        [0.00490574, 0.79481741, 0.08887725],
        [0.01482835, 0.41434266, 0.6668987 ]],

       [[0.68209787, 0.9032678 , 0.87111643],
        [0.83040063, 0.40690954, 0.43069501],
        [0.44060348, 0.6184007 , 0.43669199]],

       [[0.25732679, 0.90222633, 0.86477316],
        [0.59079063, 0.23197552, 0.73234759],
        [0.02420268, 0.58581809, 0.04223459]]])

randn Create random numbers that follow standard normal distribution.

In [23]:

vals = numpy.random.randn(5000)
vals.shape

vals = numpy.random.randn(5000) vals.shape

Out[23]:

(5000,)

In [25]:

import matplotlib.pyplot as plt
%matplotlib inline
plt.hist(vals, bins=50);

import matplotlib.pyplot as plt %matplotlib inline plt.hist(vals, bins=50);

randint¶

randint returns randomly distributed integers between specified range

In [19]:

numpy.random.randint(low=30, high=200, size=10)

numpy.random.randint(low=30, high=200, size=10)

Out[19]:

array([ 37,  43,  76,  36,  67,  45, 165,  75, 165,  40])

Array inspection¶

shape¶

Use shape property to get the dimensions

In [20]:

arr2 = numpy.random.rand(3,3,3)
arr2.shape

arr2 = numpy.random.rand(3,3,3) arr2.shape

Out[20]:

(3, 3, 3)

In [21]:

arr3 = numpy.random.randint(low=30, high=200, size=10)
arr3.shape

arr3 = numpy.random.randint(low=30, high=200, size=10) arr3.shape

Out[21]:

(10,)

thus shape is returned as a tuple.

datatype of the elements¶

In [22]:

arr2.dtype

arr2.dtype

Out[22]:

dtype('float64')

In [23]:

arr3.dtype

arr3.dtype

Out[23]:

dtype('int32')

max and min elements¶

In [25]:

arr2

arr2

Out[25]:

array([[[ 0.04539372,  0.63141413,  0.89693763],
        [ 0.55265413,  0.16925386,  0.83917698],
        [ 0.75055999,  0.26155305,  0.33921729]],

       [[ 0.35913789,  0.20122447,  0.10491535],
        [ 0.01784351,  0.20815688,  0.90825816],
        [ 0.69680734,  0.3975908 ,  0.63961161]],

       [[ 0.06461796,  0.99271516,  0.02077921],
        [ 0.26578436,  0.40538054,  0.58002467],
        [ 0.53456854,  0.85680407,  0.66601052]]])

In [24]:

arr2.max() #max element in the entire 3d array

arr2.max() #max element in the entire 3d array

Out[24]:

0.99271516081323574

In [26]:

arr2.max(axis=1) #max in each column

arr2.max(axis=1) #max in each column

Out[26]:

array([[ 0.75055999,  0.63141413,  0.89693763],
       [ 0.69680734,  0.3975908 ,  0.90825816],
       [ 0.53456854,  0.99271516,  0.66601052]])

In [32]:

arr2.max(axis=2) #max in each row

arr2.max(axis=2) #max in each row

Out[32]:

array([[ 0.89693763,  0.83917698,  0.75055999],
       [ 0.35913789,  0.90825816,  0.69680734],
       [ 0.99271516,  0.58002467,  0.85680407]])

min method to find the minimum

In [34]:

arr2.min()

arr2.min()

Out[34]:

0.017843505898512246

In [35]:

arr2.min(axis=1)

arr2.min(axis=1)

Out[35]:

array([[ 0.04539372,  0.16925386,  0.33921729],
       [ 0.01784351,  0.20122447,  0.10491535],
       [ 0.06461796,  0.40538054,  0.02077921]])

argmax and argmin to find the location of max element¶

In [36]:

arr2.argmax() #location max element in the 3D array

arr2.argmax() #location max element in the 3D array

Out[36]:

19

In [38]:

arr2.argmax(1) #indices of each max element

arr2.argmax(1) #indices of each max element

Out[38]:

array([[2, 0, 0],
       [2, 2, 1],
       [2, 0, 2]], dtype=int64)

In [39]:

arr2.argmin(1)

arr2.argmin(1)

Out[39]:

array([[0, 1, 2],
       [1, 0, 0],
       [0, 1, 0]], dtype=int64)

Array manipulation¶

Reshape arrays¶

reshape() method on an array object and send a tuple of the rows and column dimensions

In [41]:

arr3

arr3

Out[41]:

array([145, 152, 101, 130, 152,  84, 148, 160, 121, 137])

In [42]:

arr3.shape

arr3.shape

Out[42]:

(10,)

In [43]:

arr3.reshape((5,2))

arr3.reshape((5,2))

Out[43]:

array([[145, 152],
       [101, 130],
       [152,  84],
       [148, 160],
       [121, 137]])