Numpy

  • NumPy is an optimized library designed for matrix and vector computation in Python. NLP often involves representing words and sentences as vectors (e.g., word embeddings) and performing matrix operations for neural networks. Libraries like scikit-learn, PyTorch, and spaCy all depend on NumPy under the hood.

  • Documentation

Core data type

  • np.ndarray: Main data type used for representing arrays (vectors/matrices).
  • Use the constructor function np.array() to create arrays.
import numpy as np
a = np.array([1, 2, 3])  # 1D array (vector)
b = np.array([[1, 2], [3, 4]])  # 2D array (matrix)

print(a.shape) # → (3,)
print(b.shape) # → (2,2)

Element-wise arithmetic

You can perform element-wise operations using both Python operators (+, -, *, /) and NumPy functions (np.add(), np.subtract(), np.multiply(), np.divide()).

import numpy as np

x = np.array([1, 2, 3])
y = np.array([4, 5, 6])

# Addition (x + y)
result = x + y            # Same as np.add(x, y)
print(result)             # [5 7 9]

# Subtraction (x - y)
result = x - y            # np.subtract(x, y)
print(result)             # [-3 -3 -3]

# Multiplication (element-wise)
result = result * x       # np.multiply(result, x)
print(result)             # Output: [-3 -6 -9]

# Division (element-wise)
result = result / x       # np.divide(result, x)
print(result)             # Output: [-3. -3. -3.]

The * operator performs element-wise multiplication. For matrix multiplication, use np.dot()

mat1 = np.array([[1, 2],
                 [3, 4]])
mat2 = np.array([[5, 6],
                 [7, 8]])

mat3 = np.dot(mat1, mat2)
print(mat3)
# Output:
# [[19 22]
#  [43 50]]

Indexing

NumPy offers flexible indexing methods to extract and manipulate data from arrays.

import numpy as np

x = np.random.random((3, 4))  # Random matrix of shape (3, 4)

x[:]           # Selects all elements in x
  • Equivalent to copying the whole array.
x[np.array([0, 2]), :]
  • Selects the 0th and 2nd rows from x.
  • : selects all columns.
x[1, 1:3]
  • Selects from the 1st row, elements at columns 1 and 2.
  • Returns a 1D array.
x[x > 0.5]
  • Returns all elements in x that are greater than 0.5.
  • This is called boolean masking.
  • Output is a 1D array containing only the values that satisfy the condition.

Excercise

1. How do you create a 2D array with all zeros and shape (3, 4)?

A. np.zeros([3, 4])
B. np.empty([3, 4])
C. np.ones([3, 4])
D. np.array([[0]*4]*3)

2. What does x.shape return if x = np.array([[1, 2, 3], [4, 5, 6]])?

A. (3, 2)
B. (6,)
C. (2, 3)
D. (1, 6)

3. Which of the following returns the mean of a NumPy array x?

A. mean(x)
B. x.mean()
C. np.avg(x)
D. x.mean(axis=None)

4. What is the result of np.array([1, 2, 3]) * 2?

A. [2, 4, 6]
B. [1, 2, 3, 1, 2, 3]
C. [[1, 2, 3], [1, 2, 3]]
D. Error

5. Which of the following creates a 3x1 column vector?

A. np.array([[1], [2], [3]])
B. np.array([1, 2, 3])
C. np.ones((3,))
D. np.zeros((1, 3))

6. What does np.dot(a, b) compute when a and b are 1D vectors?

A. Element-wise product
B. Outer product
C. Dot (inner) product
D. Matrix multiplication

7. What is the shape of the result of np.random.random((2, 3, 4))?

A. (3, 4)
B. (2, 3, 4)
C. (4, 3, 2)
D. (2, 12)

Check answers! 1. A 2. C 3. B and D 4. A 5. A 6. C 7. B