reshape()

Virat's 12 innings arrive as one long list. But you might want them as 3 rows of 4 (one row per month), or 2 rows of 6 (first half vs second half of the season). The numbers don't change — only how they are arranged. That is exactly what reshape does.

Definition

reshape returns the same data arranged into a new shape (a new number of rows, columns or dimensions). It never adds, removes or changes values — so the total number of elements must stay the same. Whenever possible the result is a view that shares memory with the original.

• Same data — the values are read in order (row by row by default) and poured into the new shape.
• Total must match — a 12-element array can become (3, 4), (2, 6), (12, 1) or (2, 2, 3) because 3×4 = 2×6 = 12×1 = 2×2×3 = 12. It can never become (5, 3) (15 ≠ 12).
• View — the reshaped array usually points at the same memory, so changing one changes the other (see Copy vs View).

runs (12,):  [72 45 88 31 64  0 103 52 19 77 41 93]

reshape(3, 4):          reshape(2, 6):
[[ 72  45  88  31]      [[ 72  45  88  31  64   0]
 [ 64   0 103  52]       [103  52  19  77  41  93]]
 [ 19  77  41  93]]

Syntax

a.reshape(shape, order="C")           # method form: a.reshape(3, 4) or a.reshape((3, 4))
np.reshape(a, shape, order="C")       # function form
a.reshape(-1)                         # -1 = "work this dimension out for me"

Note: the method accepts the shape as separate numbers (a.reshape(3, 4)) or as a tuple (a.reshape((3, 4))). The function form needs a tuple: np.reshape(a, (3, 4)).

The rules of reshape

1. Product of the new shape = a.size. Otherwise: ValueError: cannot reshape array of size 12 into shape (5,3).
2. At most one -1. NumPy computes it as a.size ÷ (product of the other numbers); two unknowns are ambiguous → ValueError: can only specify one unknown dimension.
3. Order is preserved. Reading the new array row by row gives the same sequence as the original (with order="C").
4. Reshape does not modify a. It returns a new array object; a.shape is unchanged unless you reassign: a = a.reshape(3, 4).

Example — valid and invalid shapes

import numpy as np

runs = np.array([72, 45, 88, 31, 64, 0, 103, 52, 19, 77, 41, 93])
print("size:", runs.size, " shape:", runs.shape)

print(runs.reshape(3, 4))
print(runs.reshape(2, 2, 3).shape)
print(runs.reshape(12, 1).shape)

try:
    runs.reshape(5, 3)
except ValueError as e:
    print("ValueError:", e)
print("runs is unchanged:", runs.shape)

size: 12  shape: (12,)
[[ 72  45  88  31]
 [ 64   0 103  52]
 [ 19  77  41  93]]
(2, 2, 3)
(12, 1)
ValueError: cannot reshape array of size 12 into shape (5,3)
runs is unchanged: (12,)

How it works: (3, 4), (2, 2, 3) and (12, 1) all hold exactly 12 elements, so they succeed. (5, 3) needs 15, so NumPy refuses. The original runs keeps its shape (12,).

Example — the -1 placeholder

import numpy as np

runs = np.array([72, 45, 88, 31, 64, 0, 103, 52, 19, 77, 41, 93])

print(runs.reshape(2, -1).shape)      # 12 / 2  -> 6
print(runs.reshape(-1, 3).shape)      # 12 / 3  -> 4
print(runs.reshape(2, 3, -1).shape)   # 12 / 6  -> 2
print(runs.reshape(-1, 1).shape)      # a column vector
print(runs.reshape(3, 4).reshape(-1)) # back to 1D

try:
    runs.reshape(-1, -1)
except ValueError as e:
    print("ValueError:", e)

(2, 6)
(4, 3)
(2, 3, 2)
(12, 1)
[ 72  45  88  31  64   0 103  52  19  77  41  93]
ValueError: can only specify one unknown dimension

-1 is extremely common in real code: reshape(-1, 1) turns any 1D array into a column (needed for per-row broadcasting and for scikit-learn), and reshape(-1) flattens anything.

Example — split a season into two halves

This is the course notebook's practical use: reshape (players, matches) into (players, half, matches_per_half) to compare form.

import numpy as np

players = ["Virat", "Rohit"]
scores = np.array([
    [72, 45, 88, 31, 64,  0, 103, 52, 19, 77, 41, 93],
    [56,  0, 74, 88, 32, 45,  61, 17, 90, 38, 72,  5],
])
halves = scores.reshape(2, 2, 6)       # axis 1: 0 = first half, 1 = second half
print("shape:", halves.shape)

avg = halves.mean(axis=2)              # average inside each half
for name, (first, second) in zip(players, avg):
    trend = "Improving" if second > first else "Declining"
    print(f"{name:<6} 1st half {first:5.1f}   2nd half {second:5.1f}   {trend}")

shape: (2, 2, 6)
Virat  1st half  50.0   2nd half  64.2   Improving
Rohit  1st half  49.2   2nd half  47.2   Declining

How it works: each player's 12 scores are split row by row into 2 groups of 6. mean(axis=2) averages over the 6 matches inside each half, giving a (2, 2) table: players × halves.

Example — order="C" vs order="F"

import numpy as np

a = np.arange(1, 7)
print("C order (row by row):")
print(a.reshape(2, 3, order="C"))
print("F order (column by column):")
print(a.reshape(2, 3, order="F"))

C order (row by row):
[[1 2 3]
 [4 5 6]]
F order (column by column):
[[1 3 5]
 [2 4 6]]

How it works: "C" (named after the C language) fills the first row completely, then the next. "F" (Fortran) fills the first column first. You will almost always use the default "C"; "F" appears when data comes from MATLAB/Fortran or column-major files.

Common mistakes

• ❌ runs.reshape(5, 3) on 12 elements → ValueError: cannot reshape array of size 12 into shape (5,3). ✅ Multiply your shape out first; it must equal a.size.
• ❌ Calling runs.reshape(3, 4) and expecting runs itself to change. ✅ Reassign: runs = runs.reshape(3, 4).
• ❌ Using two -1s. ✅ Only one dimension can be unknown.
• ❌ Thinking reshape is the same as transpose. ✅ reshape(4, 3) of a (3, 4) array re-reads the values in order; .T swaps rows and columns. They give different arrays (see the next topics).
• ❌ Assuming the reshaped array is independent. ✅ It is usually a view — modifying it changes the original. Use .copy() if needed.

Comparison — ways to change shape

Interview tip

Q: What does reshape(-1, 1) do and why is it so common? A: It turns any array into a column vector of shape (n, 1), letting NumPy compute n. It's used to make a 1D array broadcast per-row against a 2D array, and because libraries like scikit-learn require 2D input of shape (samples, features).