High-Speed Scientific Computing Using NumPy

Creating NumPy ndarrays

import numpy as np

Creating 1D ndarrays

arr1D = np.array([1.1, 2.2, 3.3, 4.4, 5.5]);

Inspect the type of the array

type(arr1D)

Creating 2D ndarrays

arr2D = np.array([1, 2],[3, 4])

Creating any-dimension ndarrays

arr4D = np.array(range(16))reshape.((2, 2, 2, 2))

Creating an ndarray with np.zeros(…)

Creates an ndarray populated with all 0s

np.zeros(shape=(2, 5))

Creating an ndarray with np.ones(…)

Each value is assigned a value of 1

np.ones(shape=(2,2))

Creating an ndarray with np.identity(…)

np.identity(3)

Creating an ndarray with np.arange(…)

np.arrange(5)

Creating an ndarray with np.random.randn(…)

Generates an ndarray of specified dimensions, with each element populated with random values drawn from a standard normal distribution (mean=0, std=1)

np.random.randn(2,2)

Data types used with NumPy ndarrays

Each element in an ndarray has the same data type.

Common data types used are np.int32, np.float64, float128, and np.bool.

np.float128 is not supported on Windows.

Creating a numpy.float64 array

np.array([-1, 0, 1], dtype=np.float64)

Creating a numpy.bool array

np.array([-1, 0, 1], dtype=np.bool)

0 gets converted to False, and all other values get converted to True.

ndarrays’ dtype attribute

arr1D.dtype

Converting underlying data types of ndarray with numpy.ndarrays.astype(…)

arr1D.astype(np.int64).dtype

Indexing of ndarrays

Direct access to an ndarray’s element

arr = np.random.randn(3, 3)
arr[0]
arr[0][1]
arr[0,1]
arr[-1]

ndarray slicing

Accessing all ndarray elements after the first one

arr[1:]

Fetching all rows, starting from row 2 and columns 1 and 2

arr[1:, :2]

Slicing with negative indices

arr[1:2, -2:-1]

Slicing with no indices

arr[:][2]
arr[:][:]

Setting values of a slice to 0

arr1 = arr[1:2]
arr1[:] = 0

The operation on the arr1 slice also changed the original arr ndarray.

Create a copy of an ndarray

arr_copy = arr.copy()
arr_copy[1:2] = 1

Boolean indexing

arr = np.random.randn(3,3)
arr < 0
arr[(arr < 0)]
(arr > -1) & (arr < 1)
arr[(arr > -1) & (arr < 1)]

Indexing with arrays

arr[[0,2]]
arr[[0,2],[1]]
arr[2,0]

Basic ndarray operation

Use an arr2D ndarray

arr2D

Scalar multiplication with an ndarray

arr2D * 4

Linear combinations of ndarrays

2*arr2D + 3*arr2D

Exponentiation of ndarrays

arr2D ** 2

Addition of an ndarray with a scalar

arr2D + 10

Transposing a matrix

arr2D.transpose()

Changing the layout of an ndarray

arr2D.reshape((4,1))

arr = np.random.randn(9),reshape((3,3))

Finding the minimum value in an ndarray

np.min(arr)

Calculating the absolute value

np.abs(arr)

Calculating the mean of an ndarray

np.mean(arr)

Find the mean along the columns by specifying the axis=parameter

np.mean(arr, axis=0)
np.mean(arr, axis=1)

Finding the index of the maximum value in an ndarray

Find the location of the maximum value in the ndarray

np..argmax(arr)

Find the location of the maximum value on each row

np.argmax(arr, axis=1)

Calculating the cumulative sum of elements of an ndarray

np.cumsum(arr)

A cumulative sum is an array of a running total.

A sum is a single number.

Applying the axis=parameter to the cumsum method works similarly

np.cumsum(arr, axis=1)

Finding NaNs in an ndarray

Set the second row to np.nan

arr[1, :] = np.nan

np.isnan(arr)

Finding the truth values of x1>x2 two ndarrays

arr1 = np.random.randn(9).reshape((3,3))
arr2 = np.random.randn(9).reshape((3,3))
np.greater(arr1, arr2)
arr1 > arr2

any and all Boolean operations on ndarrays

arr_bool = (arr > -0.5) & (arr < 0.5)

Return True if any element is True and otherwise returns False

arr_bool.any()
arr_bool.any(axis=1)

Return True when all elements are True, and False otherwise

arr_bool.all()
arr_bool.all(axis=1)

Sorting ndarrays

arr1D = np.random.randn(10)

Sorting straightforward

np.sort(arr1D)

Creates an array of indices that represent the location of each element in a sorted array

np.argsort(arr1D)

Searching within ndarrays

Start with an ndarray with consecutive values

arr1 = np.array(range(1,11));
arr2 = arr1 * 1000;
cond = np.random.randn(10) > 0;

Select values from one ndarray or another, depending on the condition being True or False

np.where(cond, arr1, arr2)

File operations on ndarrays

File operations with text files

arr = np.random.randn(10)
np.savetxt('arr.csv', arr, fmt='%0.2lf', delimiter=',')
arr_new = np.loadtxt('arr.csv', delimiter=',')

File operations with binary files

np.save('arr', arr)
arr_new = np.load('arr.npy')
arr == arr_new