Flattening Multi-Dimensional Arrays (flatten vs ravel)0%

Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Flattening Multi-Dimensional Arrays (flatten vs ravel)

Beginner12 min read•Updated: Jul 11, 2026

Flattening Arrays: ravel() vs flatten()

Flattening transforms a multi-dimensional array (2D matrix, 3D image tensor, or higher-order volume) into a continuous, 1-dimensional array. This operation is ubiquitous before feeding image pixels into dense neural network layers, computing histograms, or serializing data.

NumPy provides two primary methods for this operation:

arr.ravel()
arr.flatten()

While both methods return a 1D array with identical elements, their memory allocation mechanics and performance profiles are fundamentally different.

1. The Core Distinction: View vs Copy

Visual Architecture Blueprint

+---------------+---------------------+-----------------------+--------------------+
| Method        | Memory Allocation   | Modifies Original?    | Speed Performance  |
+---------------+---------------------+-----------------------+--------------------+
| arr.ravel()   | VIEW (when possible)| YES (in-place linked) | Blazing fast (0 ms)|
| arr.flatten() | COPY (always)       | NO (completely safe)  | Slower (allocates) |
+---------------+---------------------+-----------------------+--------------------+

2. Using `ravel()`: Zero-Copy Flattening

ravel() inspects the memory strides of the array. If the array is contiguous, it creates a new 1D array header pointing directly to the existing data buffer without allocating or copying any memory:

1import numpy as np
2 
3grid = np.array([[10, 20, 30],
4                 [40, 50, 60]])
5 
6# Create a 1D representation with ravel
7raveled_view = grid.ravel()
8print("Raveled 1D array:", raveled_view)
9 
10# Mutate an element in the raveled array
11raveled_view[0] = 999
12 
13# The original array IS MODIFIED!
14print("Original grid after modifying raveled view:
15", grid)
16print("Shares memory:", np.shares_memory(grid, raveled_view))

Output:

1Raveled 1D array: [10 20 30 40 50 60]
2Original grid after modifying raveled view:
3 [[999  20  30]
4 [ 40  50  60]]
5Shares memory: True

3. Using `flatten()`: Safe Copy Allocation

flatten() is a method of ndarray that always allocates fresh memory and copies every element into the new buffer:

1matrix = np.array([[1, 2, 3],
2                   [4, 5, 6]])
3 
4# Create a flat copy
5flat_copy = matrix.flatten()
6print("Flattened array:", flat_copy)
7 
8# Mutate an element in the flattened copy
9flat_copy[0] = 777
10 
11# The original array remains completely untouched
12print("Original matrix after modifying flat copy:
13", matrix)
14print("Shares memory:", np.shares_memory(matrix, flat_copy))

Output:

1Flattened array: [1 2 3 4 5 6]
2Original matrix after modifying flat copy:
3 [[1 2 3]
4 [4 5 6]]
5Shares memory: False

4. When Does `ravel()` Fall Back to a Copy?

If an array has non-contiguous memory layout—such as after a transpose or step-sliced indexing—it is mathematically impossible to represent a flattened version with a single uniform 1D stride. In that case, ravel() automatically allocates a copy:

1base = np.arange(9).reshape(3, 3)
2transposed = base.T  # Transpose changes strides, breaking C-contiguity
3 
4print("Is contiguous?", transposed.flags.c_contiguous)  # False
5r = transposed.ravel()
6print("Shares memory after transpose?", np.shares_memory(transposed, r))  # False

5. Flattening Orders: 'C' vs 'F'

Both methods accept the order parameter:

'C': Flatten row-by-row (default).
'F': Flatten column-by-column.

1sample = np.array([[1, 2],
2                   [3, 4]])
3 
4print("Row-major (C):", sample.flatten(order='C'))  # [1, 2, 3, 4]
5print("Col-major (F):", sample.flatten(order='F'))  # [1, 3, 2, 4]

Interactive Quiz• Knowledge Check

Multiple Choice Questions

Question 1

What is the primary difference between arr.ravel() and arr.flatten()?

Question 2

If you execute b = a.flatten(); b[0] = 100, what happens to a[0, 0]?

Question 3

Under what condition will arr.ravel() return a copy instead of a view?

Question 4

Which function can be imported as a top-level NumPy function: np.ravel(a) or np.flatten(a)?

Question 5

What is the result of np.array([[10, 20], [30, 40]]).ravel(order='F')?

Next Lesson

Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Continue learning with hands-on practice, examples, and exercises in the upcoming topic.

Related Lessons

Previous Lesson	Next Lesson
Reshaping Arrays (reshape with -1 auto-dimension)	Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Practice Quiz

Test your understanding of this lesson with 5 questions. Each question has one correct answer.

Previous Next

PreviousPrevNext

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

Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Flattening Multi-Dimensional Arrays (flatten vs ravel)

Beginner12 min read•Updated: Jul 11, 2026

Study Materials Cheatsheet

Flattening Arrays: ravel() vs flatten()

NumPy provides two primary methods for this operation:

arr.ravel()
arr.flatten()

While both methods return a 1D array with identical elements, their memory allocation mechanics and performance profiles are fundamentally different.

1. The Core Distinction: View vs Copy

Visual Architecture Blueprint

+---------------+---------------------+-----------------------+--------------------+
| Method        | Memory Allocation   | Modifies Original?    | Speed Performance  |
+---------------+---------------------+-----------------------+--------------------+
| arr.ravel()   | VIEW (when possible)| YES (in-place linked) | Blazing fast (0 ms)|
| arr.flatten() | COPY (always)       | NO (completely safe)  | Slower (allocates) |
+---------------+---------------------+-----------------------+--------------------+

2. Using `ravel()`: Zero-Copy Flattening

1import numpy as np
2 
3grid = np.array([[10, 20, 30],
4                 [40, 50, 60]])
5 
6# Create a 1D representation with ravel
7raveled_view = grid.ravel()
8print("Raveled 1D array:", raveled_view)
9 
10# Mutate an element in the raveled array
11raveled_view[0] = 999
12 
13# The original array IS MODIFIED!
14print("Original grid after modifying raveled view:
15", grid)
16print("Shares memory:", np.shares_memory(grid, raveled_view))

Output:

1Raveled 1D array: [10 20 30 40 50 60]
2Original grid after modifying raveled view:
3 [[999  20  30]
4 [ 40  50  60]]
5Shares memory: True

3. Using `flatten()`: Safe Copy Allocation

flatten() is a method of ndarray that always allocates fresh memory and copies every element into the new buffer:

1matrix = np.array([[1, 2, 3],
2                   [4, 5, 6]])
3 
4# Create a flat copy
5flat_copy = matrix.flatten()
6print("Flattened array:", flat_copy)
7 
8# Mutate an element in the flattened copy
9flat_copy[0] = 777
10 
11# The original array remains completely untouched
12print("Original matrix after modifying flat copy:
13", matrix)
14print("Shares memory:", np.shares_memory(matrix, flat_copy))

Output:

1Flattened array: [1 2 3 4 5 6]
2Original matrix after modifying flat copy:
3 [[1 2 3]
4 [4 5 6]]
5Shares memory: False

4. When Does `ravel()` Fall Back to a Copy?

1base = np.arange(9).reshape(3, 3)
2transposed = base.T  # Transpose changes strides, breaking C-contiguity
3 
4print("Is contiguous?", transposed.flags.c_contiguous)  # False
5r = transposed.ravel()
6print("Shares memory after transpose?", np.shares_memory(transposed, r))  # False

5. Flattening Orders: 'C' vs 'F'

Both methods accept the order parameter:

'C': Flatten row-by-row (default).
'F': Flatten column-by-column.

1sample = np.array([[1, 2],
2                   [3, 4]])
3 
4print("Row-major (C):", sample.flatten(order='C'))  # [1, 2, 3, 4]
5print("Col-major (F):", sample.flatten(order='F'))  # [1, 3, 2, 4]

Interactive Quiz• Knowledge Check

Multiple Choice Questions

Question 1

What is the primary difference between arr.ravel() and arr.flatten()?

Question 2

If you execute b = a.flatten(); b[0] = 100, what happens to a[0, 0]?

Question 3

Under what condition will arr.ravel() return a copy instead of a view?

Question 4

Which function can be imported as a top-level NumPy function: np.ravel(a) or np.flatten(a)?

Question 5

What is the result of np.array([[10, 20], [30, 40]]).ravel(order='F')?

Next Lesson

Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Continue learning with hands-on practice, examples, and exercises in the upcoming topic.

Related Lessons

Previous Lesson	Next Lesson
Reshaping Arrays (reshape with -1 auto-dimension)	Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Practice Quiz

Test your understanding of this lesson with 5 questions. Each question has one correct answer.

Previous Next

Flattening Arrays: ravel() vs flatten()

1. The Core Distinction: View vs Copy

2. Using ravel(): Zero-Copy Flattening

3. Using flatten(): Safe Copy Allocation

4. When Does ravel() Fall Back to a Copy?

5. Flattening Orders: 'C' vs 'F'

Multiple Choice Questions

Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Related Lessons

Practice Quiz

Loading Page Content...

Flattening Arrays: ravel() vs flatten()

1. The Core Distinction: View vs Copy

2. Using ravel(): Zero-Copy Flattening

3. Using flatten(): Safe Copy Allocation

4. When Does ravel() Fall Back to a Copy?

5. Flattening Orders: 'C' vs 'F'

Multiple Choice Questions

Expanding & Squeezing Dimensions (np.newaxis, expand_dims, squeeze)

Related Lessons

Practice Quiz

2. Using `ravel()`: Zero-Copy Flattening

3. Using `flatten()`: Safe Copy Allocation

4. When Does `ravel()` Fall Back to a Copy?

2. Using `ravel()`: Zero-Copy Flattening

3. Using `flatten()`: Safe Copy Allocation

4. When Does `ravel()` Fall Back to a Copy?