vstack, hstack & column_stack

np.concatenate is powerful but you must think about axis numbers and reshape 1D arrays yourself. NumPy gives three friendlier helpers whose names describe the direction: vertical stack, horizontal stack, and column stack. These are the functions taught in the "concatenate, vstack, hstack" video, and the ones you will use most.

Definition

np.vstack stacks arrays vertically (row-wise, one below another); 1D arrays of length n are first treated as rows of shape (1, n). np.hstack stacks arrays horizontally (column-wise, side by side); for 1D arrays it simply joins them end to end. np.column_stack stacks 1D arrays as columns of a 2D array; 2D inputs are joined side by side like hstack.

• vstack = concatenate along axis 0, after promoting 1D inputs to 2D rows.
• hstack = concatenate along axis 1 for 2D+ inputs, but along axis 0 for 1D inputs — that is why hstack of two 1D arrays gives a longer 1D array, not two columns.
• column_stack is the one to use when you have separate 1D arrays (runs, balls, fours) and want them as columns of a table.

Syntax

np.vstack(tup, *, dtype=None, casting="same_kind")
np.hstack(tup, *, dtype=None, casting="same_kind")
np.column_stack(tup)

Example — two players, three helpers

import numpy as np

virat = np.array([72, 45, 88, 31])
rohit = np.array([56,  0, 74, 88])

print(np.vstack((virat, rohit)))
print(np.vstack((virat, rohit)).shape)
print(np.hstack((virat, rohit)))
print(np.column_stack((virat, rohit)))

[[72 45 88 31]
 [56  0 74 88]]
(2, 4)
[72 45 88 31 56  0 74 88]
[[72 56]
 [45  0]
 [88 74]
 [31 88]]

How it works:

• vstack turned each 1D array into a row → a 2 × 4 table (players × matches).
• hstack on 1D arrays just made one long 1D array of 8 scores.
• column_stack turned each 1D array into a column → a 4 × 2 table (matches × players).

Example — add a row and a column to an existing table

squad = np.array([[72, 45, 88, 31],
                  [56,  0, 74, 88]])
hardik = np.array([45, 32, 0, 67])
print(np.vstack([squad, hardik]))          # new row

match5 = np.array([64, 32])
print(np.column_stack([squad, match5]))    # new column

[[72 45 88 31]
 [56  0 74 88]
 [45 32  0 67]]
[[72 45 88 31 64]
 [56  0 74 88 32]]

How it works: vstack accepted a 1D row next to a 2D table (it reshapes for you). column_stack turned match5 into a (2, 1) column and attached it.

Why not hstack for the column?

np.hstack([squad, match5])

ValueError: all the input arrays must have same number of dimensions, but the array at index 0 has 2 dimension(s) and the array at index 1 has 1 dimension(s)

hstack does not promote 1D to a column. Fix: np.hstack([squad, match5.reshape(-1, 1)]) — or just use column_stack.

Example — build a stats table from separate lists

runs = np.array([72, 45, 88])
balls = np.array([48, 32, 55])
print(np.column_stack((runs, balls)))

[[72 48]
 [45 32]
 [88 55]]

Each row is now one innings: [runs, balls]. (The same result as np.vstack((runs, balls)).T.)

Comparison — which one do I need?

Note: np.row_stack is an old alias of vstack. In NumPy 2 calling it emits a DeprecationWarning ("row_stack alias is deprecated. Use np.vstack directly.") — always write vstack.

Common mistakes

• ❌ np.hstack((runs, balls)) expecting two columns. ✅ That gives one long 1D array — use np.column_stack.
• ❌ np.vstack(a, b) with two separate arguments. ✅ np.vstack((a, b)) — one tuple.
• ❌ vstacking rows of different lengths. ✅ Every row must have the same number of columns.
• ❌ Using np.row_stack. ✅ Use np.vstack.

Interview tip

Q: Given 1D arrays x and y of length n, how do you make an (n, 2) array of points? A: np.column_stack((x, y)) — or np.stack((x, y), axis=1), or np.vstack((x, y)).T. All three give the same (n, 2) result.