5.1. NumPy¶
5.1.1. numpy.ravel: Flatten a NumPy Array¶
If you want to get a 1-D array of a multi-dimensional array, try numpy.ravel(arr). You can either read the elements in the same row first or read the elements in the same column first.
import numpy as np
arr = np.array([[1, 2], [3, 4]])
arr
array([[1, 2],
[3, 4]])
np.ravel(arr)
array([1, 2, 3, 4])
np.ravel(arr, order="F")
array([1, 3, 2, 4])
5.1.2. np.squeeze: Remove Axes of Length One From an Array¶
If one or more of your axes are of length one, you can remove those axes using np.squeeze.
import numpy as np
arr = np.array([[[1], [2]], [[3], [4]]])
arr
array([[[1],
[2]],
[[3],
[4]]])
arr.shape
(2, 2, 1)
new_arr = np.squeeze(arr)
new_arr
array([[1, 2],
[3, 4]])
new_arr.shape
(2, 2)
5.1.3. Use List to Change the Positions of Rows or Columns in a NumPy Array¶
If you want to change the positions of rows or columns in a NumPy array, simply use a list to specify the new positions as shown below.
arr = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
arr
array([[1, 2, 3],
[4, 5, 6],
[7, 8, 9]])
new_row_position = [1, 2, 0]
new_arr = arr[new_row_position, :]
new_arr
array([[4, 5, 6],
[7, 8, 9],
[1, 2, 3]])
5.1.4. Difference Between NumPy’s All and Any Methods¶
If you want to get the row whose ALL values satisfy a certain condition, use NumPy’s all method.
a = np.array([[1, 2, 1], [2, 2, 5]])
# get the rows whose all values are fewer than 3
mask_all = (a < 3).all(axis=1)
a[mask_all]
array([[1, 2, 1]])
To get the row whose AT LEAST one value satisfies a certain condition, use NumPy’s any method.
mask_any = (a < 3).any(axis=1)
a[mask_any]
array([[1, 2, 1],
[2, 2, 5]])
5.1.5. Double numpy.argsort: Get Rank of Values in an Array¶
If you want to get the index of the sorted list for the original list, apply numpy.argsort() twice.
a = np.array([2, 1, 4, 7, 3])
# Get rank of values in an array
a.argsort().argsort()
array([1, 0, 3, 4, 2])
In the example above, 1 is the smallest value so it is indexed 0. 2 is the second-largest value to it is indexed 1.
5.1.6. Get the Index of the Max Value in a NumPy Array¶
To get the index of the max value in a NumPy array, use np.argmax. This can be helpful to get the highest probability in an array of probabilities.
a = np.array([0.2, 0.4, 0.7, 0.3])
np.argmax(a)
2
5.1.7. np.where: Replace Elements of a NumPy Array Based on a Condition¶
If you want to replace elements of a NumPy array based on a condition, use numpy.where.
arr = np.array([[1, 4, 10, 15], [2, 3, 8, 9]])
# Multiply values that are less than 5 by 2
np.where(arr < 5, arr * 2, arr)
array([[ 2, 8, 10, 15],
[ 4, 6, 8, 9]])
5.1.8. array-to-latex: Turn a NumPy Array into Latex¶
!pip install array-to-latex
Sometimes you might want to use latex to write math. You can turn a NumPy array into latex using array-to-latex.
import array_to_latex as a2l
a = np.array([[1, 2, 3], [4, 5, 6]])
latex = a2l.to_ltx(a)
latex
\begin{bmatrix}
1.00 & 2.00 & 3.00\\
4.00 & 5.00 & 6.00
\end{bmatrix}
I copied and pasted the output of array-to-latex to the Markdown cell of Jupyter Notebook, and below is the output.
\begin{bmatrix} 1.00 & 2.00 & 3.00\ 4.00 & 5.00 & 6.00 \end{bmatrix}
5.1.9. NumPy Comparison Operators¶
If you want to get elements of a NumPy array that are greater, smaller, or equal to a value or an array, simply use comparison operators such as <, <=, >, >=, ==.
a = np.array([1, 2, 3])
b = np.array([4, 1, 2])
a < 2
array([ True, False, False])
a < b
array([ True, False, False])
a[a < b]
array([1])
5.1.10. NumPy.linspace: Get Evenly Spaced Numbers Over a Specific Interval¶
If you want to get evenly spaced numbers over a specific interval, use numpy.linspace(start, stop, num). The code below shows a use case of the numpy.linspace method.
import matplotlib.pyplot as plt
x = np.linspace(2, 4, num=10)
x
array([2. , 2.22222222, 2.44444444, 2.66666667, 2.88888889,
3.11111111, 3.33333333, 3.55555556, 3.77777778, 4. ])
y = np.arange(10)
plt.plot(x, y)
plt.show()
5.1.11. NumPy.testing.assert_almost_equal: Check if Two Arrays Are Equal up to a Certain Precision¶
Sometimes, you might only want to check if two arrays are equal up to a certain precision. If so, use numpy.testing.assert_almost_equal.
from numpy.testing import assert_almost_equal, assert_array_equal
a = np.array([[1.222, 2.222], [3.222, 4.222]])
test = np.array([[1.221, 2.221], [3.221, 4.221]])
assert_almost_equal(a, test, decimal=2)
assert_array_equal(a, test)
---------------------------------------------------------------------------
AssertionError Traceback (most recent call last)
/tmp/ipykernel_58623/1850860365.py in <module>
5 assert_almost_equal(a, test, decimal=2)
6
----> 7 assert_array_equal(a, test)
[... skipping hidden 1 frame]
~/book/venv/lib/python3.8/site-packages/numpy/testing/_private/utils.py in assert_array_compare(comparison, x, y, err_msg, verbose, header, precision, equal_nan, equal_inf)
842 verbose=verbose, header=header,
843 names=('x', 'y'), precision=precision)
--> 844 raise AssertionError(msg)
845 except ValueError:
846 import traceback
AssertionError:
Arrays are not equal
Mismatched elements: 4 / 4 (100%)
Max absolute difference: 0.001
Max relative difference: 0.000819
x: array([[1.222, 2.222],
[3.222, 4.222]])
y: array([[1.221, 2.221],
[3.221, 4.221]])