Numpy Cheat Sheet
Python Package
SLICING (INDEXING/SUBSETTING)
numPy (numerical Python)
What is NumPy?
Foundation package for scientific computing in Python
Why NumPy?
•Numpy ‘ndarray’ is a much more efficient way
of storing and manipulating “numerical data”
than the built-in Python data structures.
•Libraries written in lower-level languages, such
as C, can operate on data stored in Numpy
‘ndarray’ without copying any data.
N-DIMENSIONAL ARRAY (NDARRAY)
What is NdArray?
Fast and space-efficient multidimensional array
(container for homogeneous data) providing vectorized
arithmetic operations
Create NdArray np.array(seq1)
# seq1 - is any sequence like object,
i.e. [1, 2, 3]
Create Special
NdArray
1, np.zeros(10)
# one dimensional ndarray with 10
elements of value 0
2, np.ones(2, 3)
# two dimensional ndarray with 6
elements of value 1
3, np.empty(3, 4, 5) *
# three dimensional ndarray of
uninitialized values
4, np.eye(N) or
np.identity(N)
# creates N by N identity matrix
NdArray version of
Python’s range
np.arange(1, 10)
Get # of Dimension ndarray1.ndim
Get Dimension Size dim1size, dim2size, .. =
ndarray1.shape
Get Data Type ** ndarray1.dtype
Explicit Casting ndarray2 = ndarray1.
astype(np.int32) ***
*Cannot assume empty() will return all zeros.
It could be garbage values.
**
Default data type is ‘np.float64’. This is
equivalent to Python’s float type which is 8
bytes (64 bits); thus the name ‘float64’.
*** If casting were to fail for some reason,
‘TypeError’ will be raised.
SLICING (INDEXING/SUBSETTING)
•Slicing (i.e. ndarray1[2:6]) is a ‘view’ on
the original array. Data is NOT copied. Any
modifications (i.e. ndarray1[2:6] = 8) to the
‘view’ will be reflected in the original array.
•Instead of a ‘view’, explicit copy of slicing via :
ndarray1[2:6].copy()
•Multidimensional array indexing notation :
ndarray1[0][2] or ndarray1[0, 2]
* Boolean indexing :
ndarray1[(names == ‘Bob’) | (names ==
‘Will’), 2:]
# ‘2:’ means select from 3rd column on
*Selecting data by boolean indexing
ALWAYS creates a copy of the data.
*The ‘and’ and ‘or’ keywords do NOT work
with boolean arrays. Use & and |.
* Fancy indexing (aka ‘indexing using integer arrays’)
Select a subset of rows in a particular order :
ndarray1[ [3, 8, 4] ]
ndarray1[ [-1, 6] ]
# negative indices select rows from the end
*Fancy indexing ALWAYS creates a
copy of the data.
Created by Arianne Colton and Sean Chen
Based on content from
‘Python for Data Analysis’ by Wes McKinney
Updated: August 18, 2016
Setting data with assignment :
ndarray1[ndarray1 < 0] = 0 *
* If ndarray1 is two-dimensions, ndarray1 < 0
creates a two-dimensional boolean array.
numPy (numerical Python)
5. Boolean arrays methods
Count # of ‘Trues’
in boolean array
(ndarray1 > 0).sum()
If at least one
value is ‘True’
ndarray1.any()
If all values are
‘True’
ndarray1.all()
Note: These methods also work with non-boolean
arrays, where non-zero elements evaluate to True.
6. Sorting
Inplace sorting ndarray1.sort()
Return a sorted
copy instead of
inplace
sorted1 =
np.sort(ndarray1)
7. Set methods
Return sorted
unique values
np.unique(ndarray1)
Test membership
of ndarray1 values
in [2, 3, 6]
resultBooleanArray =
np.in1d(ndarray1, [2,
3, 6])
•Other set methods : intersect1d(), union1d(),
setdiff1d(), setxor1d()
8. Random number generation (np.random)
•Supplements the built-in Python random * with
functions for efficiently generating whole arrays
of sample values from many kinds of probability
distributions.
samples = np.random.normal(size =(3, 3))
*Python built-in random ONLY samples
one value at a time.
COMMON OPERATIONS
1. Transposing
•A special form of reshaping which returns a ‘view’
on the underlying data without copying anything.
ndarray1.transpose() or
ndarray1.T or
ndarray1.swapaxes(0, 1)
2. Vectorized wrappers (for functions that
take scalar values)
•math.sqrt() works on only a scalar
np.sqrt(seq1) # any sequence (list,
ndarray, etc) to return a ndarray
3. Vectorized expressions
•np.where(cond, x, y) is a vectorized version
of the expression ‘x if condition else y’
np.where([True, False], [1, 2],
[2, 3]) => ndarray (1, 3)
•Common Usages :
np.where(matrixArray > 0, 1, -1)
=> a new array (same shape) of 1 or -1 values
np.where(cond, 1, 0).argmax() *
=> Find the first True element
*
argmax() can be used to find the
index of the maximum element.
Example usage is find the first
element that has a “price > number”
in an array of price data.
4. Aggregations/Reductions Methods
(i.e. mean, sum, std)
Compute mean ndarray1.mean() or
np.mean(ndarray1)
Compute statistics
over axis *
ndarray1.mean(axis = 1)
ndarray1.sum(axis = 0)
*axis = 0 means column axis, 1 is row axis.