NumPy: Numerical Computing Library

Introduction
NumPy is a powerful Python library for numerical computing, enabling efficient and fast array
operations. It is extensively used for data science, machine learning, scientific computing, and
more. This cheat sheet covers the key concepts, with practical examples of input and output,
and demonstrates various methods for different tasks, from basic array operations to
advanced machine learning applications.

Installing NumPy
You can install NumPy using either pip or conda:
• Pip:
python

pip install numpy

• Conda

python

conda install numpy

Arrays in NumPy
Array Creation:
• 1D array: It’s a simple one-dimensional array with three elements, as follows:

python

import numpy as np
arr = [Link]([1, 2, 3])
print("1D Array:", arr)

NUMPY: NUMERICAL COMPUTING LIBRARY 01

 • Input:

python

[1, 2, 3]

• Output:

python

[1 2 3]

• 2D array: It’s a two-dimensional array with 2 rows and 2 columns, as follows:

python

arr_2d = [Link]([[1, 2], [3, 4]])

print("2D Array:", arr_2d)

• Input:

python

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

• Output:

python

[[1 2]
[3 4]]

• 3D array: It’s a three-dimensional array with nested arrays, as follows:

python

arr_3d = [Link]([[[1], [2]], [[3], [4]]])

print("3D Array:", arr_3d)

NUMPY: NUMERICAL COMPUTING LIBRARY 02

 • Input:

python

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

• Output:

python

[[[1]
[2]]

[[3]
[4]]]

NumPy Basics

Basic Indexing
• Accessing elements in a 1D array: Access the first element of a 1D array.

python

arr = [Link]([1, 2, 3])

print("First Element:", arr[0])

• Input:

python

[1, 2, 3]

• Output:

python

NUMPY: NUMERICAL COMPUTING LIBRARY 03

 • Accessing elements in a 2D array: Access the element in the first row and second
column.

python

arr_2d = [Link]([[1, 2], [3, 4]])

print("Element in (0, 1):", arr_2d[0, 1])

• Input:

python

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

• Output:

python

Array Attributes
Shape, size, and data type: Get the shape, size, and data type of a 2D array.

python

print("Shape:", arr_2d.shape)
print("Size:", arr_2d.size)
print("Data Type:", arr_2d.dtype)

• Input:

python

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

NUMPY: NUMERICAL COMPUTING LIBRARY 04

 • Output:

python

Shape: (2, 2)
Size: 4
Data Type: int64

Advanced NumPy Array Operations

Array Indexing and Selection
• Subsetting with conditions: Select only positive values from the array.

python

arr = [Link]([1, -2, 3, -4])

positive_values = arr[arr > 0]
print("Positive Values:", positive_values)

• Input:

python

[1, -2, 3, -4]

• Output:

python

[1 3]

• Fancy indexing: Use an index array to select specific elements.

python

arr = [Link]([1, 2, 3, 4, 5])

selected_values = arr[[0, 2, 4]]
print("Selected Values:", selected_values)

NUMPY: NUMERICAL COMPUTING LIBRARY 05

 • Input:

python

[1, 2, 3, 4, 5]

• Output:

python

[1 3 5]

Array Manipulation
• Transpose of an array: The transpose flips the array over its diagonal.

python

arr_2d = [Link]([[1, 2], [3, 4]])

transposed = arr_2d.T
print("Transposed Array:", transposed)

• Input:

python

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

• Output:

python

[[1 3]
[2 4]]

• Reshape array: Change the shape of an array.

python

arr = [Link]([1, 2, 3, 4, 5, 6])

reshaped = [Link](2, 3)
print("Reshaped Array:", reshaped)

NUMPY: NUMERICAL COMPUTING LIBRARY 06

 • Input:

python

[1, 2, 3, 4, 5, 6]

• Output:

python

[[1 2 3]
[4 5 6]]

Broadcasting in NumPy

What Is Broadcasting?
Broadcasting allows NumPy to perform operations on arrays of different shapes by
automatically expanding the smaller array. Here’s a visual aid to help explain the concept:
Diagram:

x (3 x 3) y (3) Result (3 x 3)

0 0 0 20 21 22 20 21 22

1 2 3 + 20 21 22 = 21 23 25

4 5 6 20 21 22 24 26 28

python

import numpy as np

# Example 1: Broadcasting a 1D array to a 2D array

a = [Link]([1, 2, 3])
b = [Link]([[1], [2]])

Code continues on the next page...

NUMPY: NUMERICAL COMPUTING LIBRARY 07

 python

# Broadcasting a to match the shape of b

a_broadcasted = [Link](a, (2, 1)) # Repeat 'a' twice along axis 0 (rows)

# Calculate result
result = a_broadcasted + b
print("Array a (1D): ", a)
print("Array b (2D): ", b)
print("After Broadcasting:")
print("Array a:", a_broadcasted)
print("Array b:", b)
print("Result: ", result)

#Example 2: Broadcasting a 1D array with a scalar

scalar = 5
result = a + scalar

print("\nScalar value:", scalar)

print("Array a:", a)
print("Result of broadcasting a scalar and an array: ", result)

• Output:

python

Result: [6 7 8]

Broadcasting Example
• Adding Arrays of Different Shapes:

python

a = [Link]([1, 2, 3])
b = [Link]([[1], [2]])
result = a + b
print("Broadcasted Result:", result)

• Input:

python

a = [1, 2, 3]
b = [[1], [2]]

NUMPY: NUMERICAL COMPUTING LIBRARY 08

 • Output:

python

[[2 3 4]
[3 4 5]]

NumPy for Statistics

Normal Distribution
• Generate random numbers from a normal distribution.

python

mean, std_dev, size = 0, 1, 1000

data = [Link](mean, std_dev, size)
print("Normal Distribution Sample:", data[:5])

• Output

python

Normal Distribution Sample: [ 0.21823085 -0.21952287 -1.13819249 -0.43916598 0.02013997]

Standard Deviation
• Compute the standard deviation.

python

print("Standard Deviation:", [Link](data)) # Example: ~1.0

• Output

python

Output: Standard Deviation: 1.0203029981828742

Code continues on the next page...

NUMPY: NUMERICAL COMPUTING LIBRARY 09

 Histogram Visualization
• Use [Link] to get bin counts.
python

counts, bins = [Link](data, bins=10)

print("Histogram Counts:", counts)
print("Histogram Bins:", bins)

• Output

python

Histogram Counts: [ 3 14 55 146 243 257 176 78 18 10]

Histogram Bins: [-3.52374524 -2.84559269 -2.16744015 -1.4892876 -0.81113505 -0.13298251 0.54517004 1.22332258
1.90147513 2.57962768 3.25778022]

Mean and Axis-Specific Mean

• Compute the overall mean.

python

print("Mean:", [Link](data)) # Example: ~0.0

Mean: -0.040803288331032334

• Compute mean along a specific axis.

python

arr = [Link]([[1, 2, 3], [4, 5, 6]])

print("Mean (axis=0):", [Link](arr, axis=0)) # [2.5, 3.5, 4.5]

Mean (axis=0): [2.5 3.5 4.5]

NUMPY: NUMERICAL COMPUTING LIBRARY 10

 Percentiles and Quartiles
• Compute percentiles.

python

print("50th Percentile:", [Link](data, 50)) # Median

50th Percentile: -0.020241045615570108

• Compute quartiles.

python

print("25th and 75th Percentiles:", [Link](data, [25, 75])) # [Q1, Q3]

25th and 75th Percentiles: [-0.71818199 0.61807325]

NumPy for Data Scientists

NumPy: I/O and Data Handling

I/O: Input and Output

Save to the disk:

• Save a NumPy array to a .npy file.

python

import numpy as np

arr = [Link]([1, 2, 3, 4, 5])

[Link]('array_file.npy', arr)
print("Array saved to 'array_file.npy'")

Load from the disk

• Load a NumPy array to a .npy file.

python

loaded_arr = [Link]('array_file.npy')
print("Loaded Array:", loaded_arr)

NUMPY: NUMERICAL COMPUTING LIBRARY 11

 Save/Load text:
• Save a NumPy array to a text file.

python

[Link]('array_file.txt', arr, fmt='%d')

print("Array saved to 'array_file.txt'")

• Load the array back from the text file.

python

loaded_text_arr = [Link]('array_file.txt', dtype=int)

print("Loaded Text Array:", loaded_text_arr)

Summarize the Data

Sum with axis:
• Compute the sum across specific axes.

python

arr_2d = [Link]([[1, 2, 3], [4, 5, 6]])

print("Sum (Axis=0):", [Link](arr_2d, axis=0)) # [5, 7, 9]
print("Sum (Axis=1):", [Link](arr_2d, axis=1)) # [6, 15]

Handling the Missing Data

Replace NaN:
• Replace NaN values with a specified number.

python

arr_with_nan = [Link]([1, [Link], 2, [Link], 3])

cleaned_arr = np.nan_to_num(arr_with_nan, nan=0.0)
print("Original Array with NaN:", arr_with_nan)
print("Cleaned Array:", cleaned_arr)

NUMPY: NUMERICAL COMPUTING LIBRARY 12

 Group the Data
Group mean:
• Compute group means using slicing or indexing.

python

data = [Link]([10, 20, 30, 40, 50])

groups = [Link]([0, 0, 1, 1, 1]) # Group labels
group_means = [data[groups == g].mean() for g in [Link](groups)]
print("Group Means:", group_means)

Sliding Windows
Create sliding windows:
• Generate sliding windows using slicing.

python

arr = [Link]([1, 2, 3, 4, 5, 6])

window_size = 3
sliding_windows = [Link]([arr[i:i + window_size] for i in range(len(arr) - window_size + 1)])
print("Sliding Windows:\n", sliding_windows)

• Output

python

[[1 2 3]
[2 3 4]
[3 4 5]
[4 5 6]]

Histograms
A histogram is a graphical representation of the distribution of data.
• 1D histogram: Plot a histogram for a 1D array of random data.

python

import [Link] as plt

data = [Link](1000)
Code continues on the next page...

NUMPY: NUMERICAL COMPUTING LIBRARY 13

 python

[Link](data, bins=30, alpha=0.7)

[Link]('1D Histogram')
[Link]()

• Input:

python

[Link](1000)

• Output: A histogram plot representing the distribution of values in the data.

1D Histogram

80

60

40

20

0
-3 -2 -1 0 1 2 3

• 2D histogram: Plot a histogram for a 1D array of random data.

python

x = [Link](1000)
y = [Link](1000)
plt.hist2d(x, y, bins=30, cmap='Blues')
[Link]('2D Histogram')
[Link]()
[Link]()

NUMPY: NUMERICAL COMPUTING LIBRARY 14

 • Input:

python

[Link](1000)
[Link](1000)

• Output: A 2D histogram plot showing the relationship between x and y.

2D Histogram
12
3

10
2

8
1

0 6

-1 4

-2 2

-3 0
-2 -1 0 1 2

NUMPY: NUMERICAL COMPUTING LIBRARY 15

 NumPy for Machine Learning Researchers
Linear Algebra
• Matrix multiplication:

python

A = [Link]([[1, 2], [3, 4]])

B = [Link]([[5, 6], [7, 8]])
result = [Link](A, B)
print("Matrix Multiplication Result:", result)

• Input:

python

A = [[1, 2], [3, 4]]

B = [[5, 6], [7, 8]]

• Output:

python

Matrix Multiplication Result: [[19 22]

[43 50]]

• Solving linear equations: Solve a system of linear equations using [Link]().

python

A = [Link]([[3, 1], [1, 2]])

b = [Link]([9, 8])
x = [Link](A, b)
print("Solution of the System:", x)

• Input:

python

A = [[3, 1], [1, 2]]

b = [9, 8]

NUMPY: NUMERICAL COMPUTING LIBRARY 16

 • Output:

python

Solution of the System: [2. 3.]

Common Mistakes
• Shape mismatch in operations:
◦ Ensure array dimensions align for operations like addition or multiplication.
◦ Example:

python

arr1 = [Link]([1, 2, 3])

arr2 = [Link]([[1], [2], [3]])
result = arr1 + arr2 # Works because of broadcasting

• Unintentional data type casting:

• Pay attention to NumPy’s default types (e.g., integers in division can truncate).

python

arr = [Link]([1, 2, 3])

print(arr / 2) # Automatically converts to float

• Using built-in Python lists instead of NumPy arrays:

◦ Mixing Python lists with NumPy arrays can cause unexpected errors.

python

arr = [Link]([1, 2, 3])

print(arr + [4, 5, 6]) # Works because of broadcasting

• Confusing axis definitions:

◦ Remember, axis=0 refers to rows (vertical operations), while axis=1 refers to columns
(horizontal operations).

NUMPY: NUMERICAL COMPUTING LIBRARY 17

 Troubleshooting Tips
• Debugging array shapes:
◦ Use [Link] to inspect dimensions.
◦ Use [Link]() to fix shape issues.

• Check for missing values:

◦ Use [Link](arr).any() to check for NaN values.

• Inspect data types:

◦ Use [Link] to confirm the array type.

• Visualize intermediate results:

◦ Print arrays or use matplotlib to plot results for verification.

Key Takeaways
• NumPy is a versatile library for numerical computations, offering features like:
◦ Efficient array operations
◦ Broadcasting and array manipulation
◦ Comprehensive mathematical, statistical, and random data capabilities
◦ Integration with visualization libraries like Matplotlib

Next Steps
• Explore advanced topics:
◦ Linear algebra
◦ Fourier transforms
◦ Statistical modeling

• Dive into related libraries:

◦ pandas: For data manipulation and analysis
◦ Matplotlib: For advanced data visualization
◦ scikit-learn: For machine learning applications

• Practice with real-world projects:

◦ Work on datasets and apply NumPy concepts to solidify learning.

NUMPY: NUMERICAL COMPUTING LIBRARY 18