Show Menu
Cheatography

Python For Data Science - NumPy Library Cheat Sheet (DRAFT) by

Python For Data Science

This is a draft cheat sheet. It is a work in progress and is not finished yet.

Numpy 1D Arrays

Code
Explan­ation & OUTPUT
import numpy as np
np is simply an alias
array_1d = np.arr­ay([2, 4, 5, 6, 7, 9])
Creating a 1-D array using a list.
np.array() takes in a list or a tuple as argument, and converts into an array
print(­arr­ay_1d)
[2 4 5 6 7 9]
print(­typ­e(a­rra­y_1d))
<class 'numpy.nd­arr­ay'>
np.arr­ay(­[it­erator, dtype)
np.arr­ay([1, 2, 3, 4], dtype=­'fl­oat32')
explicitly set the data type
array([ 1., 2., 3., 4.], dtype=­flo­at32)
array_­fro­m_list = np.arr­ay([2, 5, 6, 7])
array_­fro­m_tuple = np.arr­ay((4, 5, 8, 9))
Convert lists or tuples to arrays
np.array(2, 5, 6, 7) will throw an error
list_1 = [3, 6, 7, 5]
list_2 = [4, 5, 1, 7]
array_1 = np.array(list_1)
array_2 = np.arr­ay(­list_2)
Create two 1-D arrays
array_3 = array_­1*a­rray_2
Multiple each element of array 1 with array2
OUTPUT:[12, 30, 7, 35]
array_4 = array_1 ** 2
[9,36,­49,25] # no loop required
np.arr­ay(­[3.14, 4, 2, 3])
Unlike Python lists, NumPy is constr­ained to arrays that all contain the same type
array([ 3.14, 4. , 2. , 3. ])
If types do not match, NumPy will upcast if possible e.g. int upcasted to float

NumPy Multi-­Dim­ens­ional Arrays

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
In NumPy, dimensions are called axes
array_2d = np.arr­ay([[2, 3, 4], [5, 8, 7]])
Creating a 2-D array using two lists
axis = 0 refers to the rows
axis = 1 refers to the columns
print(array_2d)
Note arrays dont have commas unlike lists on printing
[[2 3 4]
  [5 8 7]]
np.ones((row_count,column,count),datatype)
Default is float
Create array of 1s
np.one­s((5, 3))
2D array of axes(5 x 3 )of ones


np.one­s((5, 3),dty­pe=int)
array([[1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.]])
array([[1, 1, 1],
       [1, 1, 1],
       [1, 1, 1],
       [1, 1, 1],
      [1, 1, 1]])
np.zeros((row_count,column,count),datatype):
Default is float
Create array of 0s

np.zeros(4, dtype = np.int)

np.zeros((4,2,2), dtype = np.int)
array([0, 0, 0, 0])
array([[[0, 0],
       [0, 0]],

       [[0, 0],
       [0, 0]],

       [[0, 0],
       [0, 0]],

       [[0, 0],
       [0, 0]]])
np.random.random():
Default is float
Create array of random numbers

np.random.random([3, 4])
array(­[[9.53­309­987­e-01, 7.6100­524­1e-04, 4.1197­873­9e-01, 4.54277232e-01],
   [6.878­425­77e-01, 9.0296­550­9e-01, 5.3213­908­1e-01, 5.41951709e-01],
   [8.581­887­84e-01, 1.1137­526­7e-01, 8.1163­897­0e-05, 7.0412­102­0e-­01]])
np.ara­nge­(st­art­,st­op,­ste­p,d­type):
similar to range
If dtype is not given, infer the data type from other input arguments
Create array with increments of a fixed step size

numbers = np.ara­nge(10, 100, 5)
[10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95]
np.lin­spa­ce(­sta­rt,­sto­p,n­umber of elemen­ts,­dtype):
Default is float
Create array of fixed length­;So­met­imes, you know the length of the array, not the step size

np.linspace(10, 100, 19,dty­pe=int)
array([ 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, 100])
np.full((dimensions),element_to_be_filled)
Default is int
Creating a 4 x 3 array of 7s using np.full(); default is int
np.full((4,3), 7)
array([[7, 7, 7],
     [7, 7, 7],
     [7, 7, 7],
     [7, 7, 7]])
np.til­e(a­rr,­rep­eat­_count)
Default is int
creates a new array by repeating the given array for the given repeated count
arr = ([0, 1, 2])
np.tile(arr, 3)
np.tile(arr, (3,2))
array([0, 1, 2, 0, 1, 2, 0, 1, 2])
array([[0, 1, 2, 0, 1, 2],
     [0, 1, 2, 0, 1, 2],
     [0, 1, 2, 0, 1, 2]])
np.eye­(id­ent­ity­_ma­tri­x_e­lem­ent­,dtype)
Default type is float
Create a 3 x 3 identity matrix
np.eye(3, dtype = int)
array([[1, 0, 0],
     [0, 1, 0],
     [0, 0, 1]])
np.ran­dom.ra­ndi­nt(­sta­rt,­sto­p,(­dim­ens­ions))
Only integers
rand_a­rra­y=n­p.r­and­om.r­an­dint(0, 10, (4,4))
array([[9, 3, 6, 0],
     [1, 5, 7, 5],
     [4, 2, 6, 4],
     [5, 3, 4, 6]])
Print the second row
print(­ran­d_a­rray[1, ])
[1, 5, 7, 5]
np.emp­ty(3)
Create an uninit­ialized array of three integers
array([ 1., 1., 1.]) Could be anything from memory
 
array_1d = np.arange(10)
print(array_1d)
[0 1 2 3 4 5 6 7 8 9]
Iteration of 1D array is similar to lists
for i in array_1d:
   print(­i**2)
0
1
4
9
Iterating on 2-D arrays is done with respect to the first axis (which is row, the second axis is column)
for row in array_2d: print(row)
[2 3 4]
[5 8 7]
3D array
array_3d = np.ara­nge­(24­).r­esh­ape(2, 3, 4)
print(­arr­ay_3d)
[[[ 0 1 2 3]
   [ 4 5 6 7]
   [ 8 9 10 11]]

[[12 13 14 15]
   [16 17 18 19]
   [20 21 22 23]]]
Iterating over 3-D arrays: Done with respect to the first axis
for row in array_3d:
   print(row)
[[[ 0 1 2 3]
   [ 4 5 6 7]
   [ 8 9 10 11]]
[[12 13 14 15]
   [16 17 18 19]
   [20 21 22 23]]]

NumPy Array Attributes

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
array.s­hape
Returns Shape of array in the form (n x m)
print(­"­Shape: {}".f­or­mat­(ra­nd_­arr­ay.s­hape))
Shape: (4, 4)
array.ndim
Returns number of dimensions (or axes)
print(­"­Dim­ens­ions: {}".f­or­mat­(ra­nd_­arr­ay.n­dim))
Dimens­ions: 2
array.d­type
Returns data type (int, float etc.)
print(­"­dtype: {}".f­or­mat­(ra­nd_­arr­ay.d­type))
dtype: int32
array.size
Returns total number of elements in the array
print(­"­Size: ", rand_a­rra­y.size)
Size: 16
array.i­te­msize
Returns Memory used by each array element in bytes
print(­"Item size: {}".f­or­mat­(ra­nd_­arr­ay.i­te­msize))
Item size: 4
array.n­bytes
Returns the total size (in bytes) of the array
print(­"­nby­tes­:", rand_a­rra­y.n­bytes, "­byt­es")
nbytes: 64 bytes

NumPy Array Indexing

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
Access single elements: x[start:stop:step]
Print third element
print(­arr­ay_­1d[2])
2
Access single elements: x[start:stop:step]
Print last element
print(­arr­ay_­1d[-1])
9
Access single element in a 2D array
Prints second row third column
print(­arr­ay_­2d[1, 2])
8
Access single element in a 2D array
Prints second row last column
print(­arr­ay_­2d[1, -1])
7

NumPy Array Slicing

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
Accessing subarrays
Return specific elements; Index has to be a list
x[start:stop:step]
print(array_1d[[2, 5, 6]])
print(array_1d[2, 5, 6])
Default ­start: 0, stop: size of dimension, step = 1
[2 5 6]
using [] instead of[[]] Will throw error
 
array_1d = np.arange(10)
print(array_1d)
[0 1 2 3 4 5 6 7 8 9]
Slice third element onwards
print(­arr­ay_­1d[2:])
[2 3 4 5 6 7 8 9]
Slice first three elements
print(­arr­ay_­1d[:3])
[0 1 2]
Slice third to seventh elements
print(­arr­ay_­1d[­2:7])
[2 3 4 5 6]
Subset starting 0 at increment of 2
print(­arr­ay_­1d[­0::2])
[0 2 4 6 8]
Slicing a 2D array
print(­arr­ay_­2d[1, :])
[5 8 7]
Slicing a 2D array returns an array
print(­typ­e(a­rra­y_2d[1, :]))
<class 'numpy.nd­arr­ay'>
Slicing all rows and the third column
print(­arr­ay_­2d[:, 2])
[4 7]
Slicing all rows and the first three columns
print(­arr­ay_­2d[:, :3])
[[2 3 4]
   [5 8 7]]
Slicing elements within range with step size
print(­arr­ay_­1d[­2:7:2])
[2 4 6]
import numpy as np
arr1 = np.array([1,2,3,4])
print(arr1)
arr2 = arr1[1:]
print(arr2)
arr2[1] = 8
print(arr1)
print(­arr2)
Numpy array on slicing will not return new copy.
Numpy array slicing will only return a view or reference to original array(like shallow copy)
[1 2 3 4]
[2 3 4]
[1 2 8 4]
[2 8 4]
list1 = [1,2,3,4]
print(list1)
list2 = list1[1:]
print(list2)
list2[1] = 8
print(list1)
print(­list2)
Lists on slicing will create new copy.
[1, 2, 3, 4]
[2, 3, 4]
[1, 2, 3, 4]
[2, 8, 4]

Reshaping of NumPy Arrays

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
array.r­es­hap­e(d­ime­nsions)
Default is int
x = np.arange(24)
print(x)
gives [ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23]

print(x.reshape(3, 2, 4))
3D array created from 1D array using reshape()

The last axis has 4 elements, and is printed from left to right.
The second last has 3, and is printed top to bottom
* The other axis has 2, and is printed in the two separated blocks
[[[ 0 1 2 3]
   [ 4 5 6 7]]

  [[ 8 9 10 11]
   [12 13 14 15]]

  [[16 17 18 19]
   [20 21 22 23]]]
array[­np.n­ew­axis,:] creates a row vector
array[­:,n­p.n­ewaxis] creates a column vector
print(x[np.newaxis,:])
print(­x[:­,np.ne­waxis])
[[ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23]]
[[ 0]
[ 1]
[ 2]
.... [ 22]
[ 23]]

NumPy Array Concat­enation

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
np.con­cat­ena­te(­[ar­ray1, 2,...]­,ax­is=0)
Default is along x-axis
Along x-axis means adding rows(row-wise)(axis=0)
Along y-axis means adding column­s(x­olu­mn-­wis­e)(­axis=1)
x = np.arr­ay([1, 2, 3])
y = np.arr­ay([3, 2, 1])
z = [99, 99, 99]
print(­np.c­on­cat­ena­te([x, y, z]))
The dimensions should match on the axis the arrays are being concatenated.
Here x dimensions are 1 x 3
y dimensions are 1 x 3
z dimensions are 1 x 3
Since columns are same, they can be concat­enated across x-axis
[ 1 2 3 3 2 1 99 99 99]
Concat­enate on 2D arrays with same dimensions
grid = np.arr­ay([[1, 2, 3],
         [4, 5, 6]])
np.con­cat­ena­te(­[grid, grid])
np.con­cat­ena­te(­[grid, grid], axis=1)
array([[1, 2, 3],
     [4, 5, 6],
   [1, 2, 3],
   [4, 5, 6]])
array([[1, 2, 3, 1, 2, 3],
   [4, 5, 6, 4, 5, 6]])
Concat­enate on arrays with different dimensions
np.vstack([arrays])
np.hstack([arrays])
np.dstack([arrays])
np.vst­ack­([a­rray1, array2])
array1 and 2 should have same column size
print(­np.v­st­ack­([x­,gr­id]))
[[1 2 3]
   [1 2 3]
   [4 5 6]]
np.hst­ack­([a­rray1, array2])
array1 and 2 should have same row size
print(np.hstack([x,y]))
y = np.array([[99],[99]])
np.hst­ack­([grid, y])
[1 2 3 3 2 1]
[[ 1 2 3 99]
  [ 4 5 6 99]]
np.dst­ack­([a­rrays])
same as **np.c­onc­ate­nat­e([­arr­ays­],a­xis=2)
will stack arrays along the third axis

NumPy Array Splitting

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
np.spl­it(­array, [split indices as list])
x = [1, 2, 3, 99, 99, 3, 2, 1]
x1, x2, x3 = np.spl­it(x, [3, 5])
print(x1, x2, x3)
Splitting is opposite of Concat­ena­tion.
N slice points mentioned will give N+1 arrays
[1 2 3] [99 99] [3 2 1]
 
grid = np.array([1,2,3,4,5,6,7,8,9]).reshape((3,3))
print(­grid)
vsplit only works on arrays of 2 or more dimensions
upper, lower = np.vsp­lit­(grid, [2])
print(upper)
print(­lower)
[[1 2 3]
  [4 5 6]]

[[7 8 9]]
 
left, middle, right = np.hsp­lit­(grid, [1,2])
print(left)
print(middle)
print(right)
[[1]
  [4]
  [7]]

[[2]
  [5]
  [8]]

[[3]
  [6]
  [9]]

NumPy Aggreg­ation functions

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
1D array Aggregate operations
m = np.arange(10)
print(m)
print(sum(m))
print(np.sum(m))
print(np.min(m))
print(np.max(m))
[0 1 2 3 4 5 6 7 8 9]
45
45
0
9
2D array aggregate vs normal functions
p = np.arange(9).reshape(3,3)
print(p)
print(sum(p))
print(np.sum(p))
print(­p.s­um())
[[0 1 2]
  [3 4 5]
  [6 7 8]]
[ 9 12 15]
36
36
 
print(p.min())
print(p.min(axis=0))
print(­p.m­in(­axi­s=1))
0
[0 1 2]
[0 3 6]

NumPy Inbuilt Aggregate Functions

Function Name
NaN-safe Version
Descri­ption
Most aggregates have a NaN-safe counte­rpart that computes the result while ignoring missing values,
np.sum
np.nansum
Compute sum of elements
np.prod
np.nanprod
Compute product of elements
np.mean
np.nanmean
Compute median of elements
np.std
np.nanstd
Compute standard deviation
np.var
np.nanvar
Compute variance
np.min
np.nanmin
Find minimum value
np.max
np.nanmax
Find maximum value
np.argmin
np.nan­argmin
Find index of minimum value
np.argmax
np.nan­argmax
Find index of maximum value
np.median
np.nan­median
Compute median of elements
np.per­centile
np.nan­per­centile
Compute rank-based statistics of elements
np.any
N/A
Evaluate whether any elements are true
np.all
N/A
Evaluate whether all elements are true

Mathem­atical Operations on NumPy Arrays

Syntax and Concepts
Example Code
Explan­ation & OUTPUT
np.sin­(ar­ray­_name)
np.cos­(ar­ray­_name)
a = np.ara­nge(1, 5)
print(np.sin(a))
print(np.cos(a))
print(np.exp(a))
print(­np.l­og(a))
[ 0.84147098 0.90929743 0.14112001 -0.7568025 ]
[ 0.54030231 -0.416­14684 -0.9899925 -0.65364362]
[ 2.71828183 7.3890561 20.085­53692 54.59815003]
[0. 0.69314718 1.09861229 1.3862­9436]
np.vec­tor­ize­(cu­sto­m_f­unc­tion)
a = np.arange(5)

f = np.vec­tor­ize­(lambda x: x+10)
print(­f(a))
[10 11 12 13 14]
Custom function on 2D array
Previous functions can be reused for 2D arrays too
b = np.lin­spa­ce(1, 100, 10,dtype=int)
print(b)
print(­f(b))
[ 1 12 23 34 45 56 67 78 89 100]
[ 11 22 33 44 55 66 77 88 99 110]
np.linalg
Applies common linear algebra operations
np.lin­alg.in­v(a­rra­y_name)
returns array of inverse of a matrix
np.lin­alg.de­t(a­rra­y_name)
returns determ­inant value of the matrix
np.lin­alg.ei­g(a­rra­y_name)
returns eigenv­alues and eigenv­ectors of the matrix
np.dot­(ar­ray­1,a­rray2))
returns matrix multip­lic­ation