NumPy Structured Arrays

Last Updated: 09 Nov 2025

Structured arrays let you store heterogeneous data (like CSV, sensor logs, database rows) in a single NumPy array — fast, memory-efficient, vectorized.

Hinglish Tip: “Structured array = ek hi array me naam, umar, salary — jaise Excel sheet!”

1. Define dtype with Named Fields

import numpy as np

# Define schema: name (string), age (int), salary (float)
dt = np.dtype([
    ('name', 'U10'),     # Unicode string, max 10 chars
    ('age', 'i4'),       # 32-bit integer
    ('salary', 'f8')     # 64-bit float
])

# Create array
employees = np.array([
    ('Amit', 25, 50000.0),
    ('Suraj',   30, 75000.0),
    ('Raj',   28, 65000.0)
], dtype=dt)

print(employees)

Output:

[('Amit', 25, 50000.) ('Suraj', 30, 75000.) ('Raj', 28, 65000.)]

2. Access by Field Name

print("Names:", employees['name'])
print("Ages:", employees['age'])
print("Salaries:", employees['salary'])

3. Vectorized Operations

# 10% bonus for all
employees['salary'] *= 1.10
print(employees['salary'])

# Filter: age > 28
print(employees[employees['age'] > 28])

4. Nested & Complex dtype

complex_dt = np.dtype([
    ('id', 'i4'),
    ('position', 'f4', (3,)),   # 3D point
    ('active', '?')
])

point = np.array([(1, [10.0, 20.0, 30.0], True)], dtype=complex_dt)
print(point['position'])

5. clip

  • Cuts values that are too small or too big.
  • arr = np.clip(arr, min_value, max_value)
import numpy as np

x = np.array([-10, -3, 0, 2, 7, 15, 100])

# 1. between 0 and 10
print(np.clip(x, 0, 10))
# → [ 0  0  0  2  7 10 10]

# 2. Only lower limit (everything below -2 becomes -2)
print(np.clip(x, -2, None))
# → [ -2  -2  -2  -2   2   7 100]

# 3. Only upper limit
print(np.clip(x, None, 8))
# → [-10  -3   0   2   7   8   8]

# 4. Clip in-place (faster, no copy)
x.clip(0, 10, out=x)   # x is now changed
print(x)
# → [ 0  0  0  2  7 10 10]

6. pad

  • Pads an array with a constant value.
  • arr = np.pad(arr, pad_width, mode='constant', constant_values=0)
  • mode can be 'constant', 'edge', 'reflect', 'symmetric'.
import numpy as np

small = np.array([[1, 2],
                  [3, 4]])
# Example 1 same padding on each side
padded = np.pad(small, pad_width=2, mode='constant', constant_values=0)
print(padded

# Example 2 – Different padding on each side
# (top, bottom, left, right)
padded = np.pad(small, pad_width=(3, 1, 2, 4), mode='constant', constant_values=9)
print(padded))

7. tile

  • use to repeat an array.
  • np.tile(your_array, (rows, cols))
import numpy as np

# 1 D
small = np.array([10, 20])
print(np.tile(small, 4))

# 2 D
block = np.array([[1, 2],
                  [3, 4]])

# Repeat: 3 times down, 2 times right
big = np.tile(block, (3, 2))
print(big)
← PreviousNext →