MSK Institute • Shikohabad
Loading Page Content...
Preparing courses, live labs, and learning resources.सामग्री लोड हो रही है, कृपया प्रतीक्षा करें...
Connecting serverPlease wait...
💻 Practical Labs🚀 Free Demo📜 Verified Certificates
Built-in Creation Routines (zeros, ones, full, empty)0%
Built-in Creation Routines (zeros, ones, full, empty)
Beginner12 min read•Updated: Jul 11, 2026
Built-in Creation Routines (zeros, ones, full, empty)
Converting Python lists is convenient for small tests, but in production machine learning and scientific simulations, allocating arrays containing millions of pre-initialized cells must be done without allocating Python objects first. NumPy provides specialized built-in memory allocation routines.
1. Allocating Zeros with np.zeros()
np.zeros() allocates a contiguous block of memory and fills every byte with 0:
import numpy as np
# 1D array of 5 float zeros
z1 = np.zeros(5)
print(z1) # [0. 0. 0. 0. 0.] (Default dtype is float64)
# 2D matrix of 3 rows and 4 columns filled with integer zeros
z2 = np.zeros((3, 4), dtype=int)
print(z2)
# [[0 0 0 0]
# [0 0 0 0]
# [0 0 0 0]]
Common Use Case: Initializing weight matrices, zero-padding convolution masks, or accumulator buffers in numerical simulations.
Click to Zoom2. Allocating Ones with np.ones()
np.ones() operates identically to zeros(), but initializes every element to 1:
# Matrix of ones with shape (2, 3)
ones_arr = np.ones((2, 3), dtype=np.float32)
print(ones_arr)
# [[1. 1. 1.]
# [1. 1. 1.]]
Common Use Case: Baseline probability priors, bias vectors in neural networks, and multiplier masks.
3. Allocating Constant Values with np.full()
When you need an array initialized with any arbitrary constant value (e.g., 7, -1, 3.14, or np.nan):
# Create a 3x3 matrix filled entirely with 99
f1 = np.full((3, 3), fill_value=99)
print(f1)
# [[99 99 99]
# [99 99 99]
# [99 99 99]]
# Matrix filled with NaN (Not a Number) for missing data tracking
nan_matrix = np.full((2, 4), fill_value=np.nan)
4. High-Speed Allocation with np.empty()
What if you are about to overwrite every single element of an array immediately with data from a disk stream or sensor? Initializing cells to 0 wastes CPU write cycles. np.empty() allocates memory without initializing its values:
# Allocates memory instantly without clearing existing RAM contents
raw_buffer = np.empty((3, 3))
print(raw_buffer) # Contains uninitialized garbage values from memory!
Important Warning
Caution with np.empty():
np.empty() does not clear memory! The array will contain unpredictable residual bits (garbage values) left behind by other applications. Never read from an empty() array before writing to it!5. The Like-Family: Matching Existing Arrays
NumPy provides companion routines that mirror the shape and dtype of an existing template array:
ref = np.array([[1, 2, 3], [4, 5, 6]], dtype=np.int32)
# Same shape (2, 3) and dtype (int32), but all zeros
z_like = np.zeros_like(ref)
# Same shape and dtype, but filled with 500
f_like = np.full_like(ref, 500)
Interactive Quiz
Multiple Choice Questions
Question 1
What is the default data type (dtype) created by 'np.zeros(5)' if no explicit dtype is provided?
Question 2
What does 'np.empty((3, 3))' return?
Question 3
Which function creates a (4, 4) matrix where every element is initialized to the number -1?
Question 4
How can you quickly create an array of ones that has the exact same shape and data type as an existing array named 'features'?
Question 5
Why should 'np.zeros()' be passed a tuple for multidimensional shapes (e.g., 'np.zeros((3, 4))') rather than two separate arguments?
Next Lesson
Generating Numerical Sequences (arange, linspace, logspace)
Continue learning with hands-on practice, examples, and exercises in the upcoming topic.
Related Lessons
Practice Quiz
Test your understanding of this lesson with 5 questions. Each question has one correct answer.