Depth & Column Stacking (dstack, column_stack, stack)0%
Splitting Arrays into Chunks (split, hsplit, vsplit)

Depth & Column Stacking (dstack, column_stack, stack)

Beginner12 min read•Updated: Jul 11, 2026
Study Materials

Depth & Column Stacking: dstack(), column_stack(), & stack()

While vstack and hstack handle basic 2D horizontal and vertical stacking, real-world data science and computer vision problems frequently require working along other axes:

  • Merging independent grayscale channels into a multi-channel RGB or RGBA image tensor.
  • Stacking 1D feature vectors side-by-side as columns in a 2D design matrix.
  • Stacking arrays along an entirely new dimension (such as stacking multiple time-series frames).

1. Column Stacking: np.column_stack()

np.column_stack() takes a sequence of 1D arrays and stacks them as columns into a single 2D matrix. This is one of the cleanest, most idiomatic ways in NumPy to assemble tabular feature matrices from separate 1D variables:

Python
import numpy as np
 
# Feature vectors
feature_age = np.array([25, 30, 45, 52])
feature_income = np.array([55000, 68000, 110000, 95000])
feature_credit = np.array([710, 680, 790, 740])
 
# Assemble into a 2D design matrix (4 samples x 3 features)
design_matrix = np.column_stack((feature_age, feature_income, feature_credit))
print("Design Matrix (Shape:", design_matrix.shape, "):
", design_matrix)

Output:

Output
Design Matrix (Shape: (4, 3) ):
[[ 25 55000 710]
[ 30 68000 680]
[ 45 110000 790]
[ 52 95000 740]]
Note
For 2D arrays, column_stack() behaves identically to hstack().

2. Depth Stacking: np.dstack()

np.dstack() stacks arrays along axis 2 (the depth axis).

This is the standard operation for assembling color images from separate 2D color channel planes:

Python
# Create three 2D color channels (e.g. 4x4 image)
red_channel = np.full((4, 4), 255, dtype=np.uint8)
green_channel = np.full((4, 4), 128, dtype=np.uint8)
blue_channel = np.zeros((4, 4), dtype=np.uint8)
 
# Stack along depth to construct an RGB image tensor of shape (4, 4, 3)
rgb_image = np.dstack((red_channel, green_channel, blue_channel))
print("RGB Image Tensor shape:", rgb_image.shape) # (4, 4, 3)
print("Pixel (0, 0) RGB values:", rgb_image[0, 0]) # [255, 128, 0]

3. Creating New Axes with np.stack()

Notice the critical conceptual difference between np.concatenate() and np.stack():

  • np.concatenate(): Joins arrays along an existing axis. Output has the same number of dimensions (ndim).
  • np.stack(): Joins arrays along a NEW axis. Output has ndim + 1 dimensions.
Python
frame1 = np.ones((10, 10))
frame2 = np.ones((10, 10)) * 2
frame3 = np.ones((10, 10)) * 3
 
# Stack 2D frames into a 3D video clip along axis 0: shape (3, 10, 10)
video_clip = np.stack((frame1, frame2, frame3), axis=0)
print("Stacked axis 0 shape:", video_clip.shape)
 
# Stack along axis -1: shape (10, 10, 3)
video_channels = np.stack((frame1, frame2, frame3), axis=-1)
print("Stacked axis -1 shape:", video_channels.shape)

4. Summary of Stacking Helpers

FunctionEffective BehaviorDimensionality ChangeTypical Use Case
np.concatenateJoins along existing axisSame ndimAppending rows or columns
np.stackJoins along brand new axisndim + 1Batching images, time-series frames
np.vstackRow-wise stackingPromotes 1D $\to$ 2DAppending sample rows
np.hstackColumn-wise for 2D, 1D concatSame or 2DHorizontal image tiling
np.column_stack1D arrays as columnsAlways 2DML design matrices
np.dstackStacks along 3rd axis (depth)Promotes to 3DRGB channel merging

Multiple Choice Questions

1. How does np.stack() differ fundamentally from np.concatenate()?

A. np.stack() runs faster because it does not allocate memory B. np.stack() joins arrays along a brand new axis, increasing dimensionality by 1 C. np.stack() can only accept two arrays D. np.stack() flattens all inputs Answer: B Explanation: While np.concatenate() joins along an already existing dimension preserving ndim, np.stack() creates and inserts a brand new dimension, increasing ndim by 1.

2. If you have three 1D arrays of shape (100,), what is the shape of the output from np.column_stack((a, b, c))?

A. (300,) B. (3, 100) C. (100, 3) D. (100, 1, 3) Answer: C Explanation: np.column_stack() takes 1D vectors and aligns each as a column in a 2D matrix, producing shape (100, 3).

3. Which function is most appropriate for combining three 2D matrices of shape (1080, 1920) representing Red, Green, and Blue into a standard image array of shape (1080, 1920, 3)?

A. np.vstack B. np.hstack C. np.dstack D. np.concatenate(..., axis=0) Answer: C Explanation: np.dstack() stacks along the third axis (depth / axis 2), turning (H, W) 2D planes into an (H, W, 3) 3D volume.

4. What is the output shape of np.stack([np.zeros((4, 5)), np.zeros((4, 5))], axis=1)?

A. (4, 2, 5) B. (2, 4, 5) C. (4, 10) D. (8, 5) Answer: A Explanation: Stacking two arrays of shape (4, 5) along axis=1 inserts the new axis of size 2 at index 1, yielding shape (4, 2, 5).

5. When using np.stack(), what condition must all input arrays satisfy?

A. They must be 1D B. They must have identical dtypes and exactly the same shape C. They must be C-contiguous D. They must have fewer than 1000 elements Answer: B Explanation: All arrays passed to np.stack() must have the exact same shape, because they are being aligned along a newly introduced axis.

Next Lesson

Splitting Arrays into Chunks (split, hsplit, vsplit)

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

Related Lessons

Previous LessonNext Lesson
Concatenating & Stacking Arrays (concatenate, vstack, hstack)Splitting Arrays into Chunks (split, hsplit, vsplit)

Practice Quiz

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

PreviousNext
PrevNext