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
Reshaping Arrays (reshape with -1 auto-dimension)0%
Reshaping Arrays (reshape with -1 auto-dimension)
Beginner12 min read•Updated: Jul 11, 2026
Array Reshaping with reshape() & -1 Dimension
In machine learning models, computer vision pipelines, and matrix algebra, manipulating the dimensions and geometric layout of arrays is a ubiquitous daily requirement. For instance, a neural network layer may require flattening a 2D image matrix of (28, 28) pixels into a 1D vector of 784 features, or batching 1000 samples into mini-batches of (32, 10, 10).
NumPy's reshape() method allows you to change the shape of an array without changing its underlying data or reallocating memory.
Click to Zoom1. The Fundamental Invariant of Reshaping
The cardinal rule of reshaping is that the total number of elements (the product of dimension sizes) must remain strictly identical before and after the reshape operation:
$$\prod_{i} \text{old\_shape}[i] = \prod_{j} \text{new\_shape}[j] = \text{total\_elements}$$
import numpy as np
# Create a 1D array of 12 elements (indices 0 to 11)
a = np.arange(12)
print("Original shape:", a.shape) # (12,)
# Reshape into a 3x4 matrix (3 * 4 = 12)
matrix_3x4 = a.reshape(3, 4)
print("3x4 Matrix:
", matrix_3x4)
# Reshape into a 2x2x3 3D tensor (2 * 2 * 3 = 12)
tensor_3d = a.reshape(2, 2, 3)
print("2x2x3 Tensor:
", tensor_3d)
Incompatible Shapes Raise ValueError
If you attempt to reshape an array of 12 elements into a shape whose product is not 12 (e.g. (3, 5) which requires 15 elements), NumPy raises an exception:
# Raises: ValueError: cannot reshape array of size 12 into shape (3,5)
try:
a.reshape(3, 5)
except ValueError as e:
print("Caught Error:", e)
2. Dynamic Dimension Deduction with -1
Calculating dimension sizes manually becomes tedious and error-prone when processing dynamic datasets with variable batch sizes. NumPy allows you to specify -1 for exactly one dimension. NumPy will automatically deduce the required size for that dimension by dividing the total element count by the product of the remaining dimensions:
raw_data = np.arange(60)
# Reshape into 5 rows, automatically deduce columns (60 / 5 = 12)
grid1 = raw_data.reshape(5, -1)
print("grid1 shape:", grid1.shape) # (5, 12)
# Reshape into 4 columns, automatically deduce rows (60 / 4 = 15)
grid2 = raw_data.reshape(-1, 4)
print("grid2 shape:", grid2.shape) # (15, 4)
# Automatically deduce depth in a 3D volume
volume = raw_data.reshape(3, 4, -1)
print("volume shape:", volume.shape) # (3, 4, 5)
Rule: You can only provide
-1 for at most one dimension. Specifying arr.reshape(-1, -1) will raise ValueError: can only specify one unknown dimension.3. Memory Layout: C-Order vs Fortran-Order
When reshaping, the order in which elements are read from and written to the dimensions matters:
- C-Style (
order='C', Default): Row-major order. The last index changes fastest (rows are filled left-to-right before moving to the next row). Matches C, C++, and Python conventions. - Fortran-Style (
order='F'): Column-major order. The first index changes fastest (columns are filled top-to-bottom before moving to the next column). Matches Fortran, MATLAB, and R conventions.
v = np.arange(1, 7)
# C-order (Default)
c_matrix = v.reshape((2, 3), order='C')
print("C-order:
", c_matrix)
# Fortran-order
f_matrix = v.reshape((2, 3), order='F')
print("Fortran-order:
", f_matrix)
Output:
C-order:
[[1 2 3]
[4 5 6]]
Fortran-order:
[[1 3 5]
[2 4 6]]
4. Does reshape() Return a View or a Copy?
In almost all standard cases where the memory buffer is contiguous, reshape() returns a VIEW of the original data. Modifying elements of the reshaped array modifies the original array:
original = np.array([1, 2, 3, 4, 5, 6])
reshaped = original.reshape(2, 3)
# Mutate an element in the reshaped view
reshaped[0, 0] = 999
print("Original array:", original)
print("Shared base:", reshaped.base is original)
Output:
Original array: [999 2 3 4 5 6]
Shared base: True
If an array has been heavily sliced or transposed such that its memory is non-contiguous,
reshape() may be forced to allocate a copy to satisfy the new stride layout.Interactive Quiz
Multiple Choice Questions
Question 1
What is the fundamental requirement when reshaping an array using
arr.reshape(new_shape)?Question 2
How many dimensions can be set to
-1 in a single reshape() call?Question 3
What is the inferred shape when calling
np.arange(24).reshape(2, -1, 3)?Question 4
What is the default memory ordering used by
reshape() if the order parameter is omitted?Question 5
In standard contiguous memory, what does
arr.reshape(...) return?Next Lesson
Flattening Multi-Dimensional Arrays (flatten vs ravel)
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.