🕯️LORE

Reshaping Arrays

Properties2
tagsnumpy, python, reshape, stacking
aliasesreshape, flatten, transpose, concatenate

3 min read

Reshaping Arrays

Definition

Reshaping tools change an array’s shape or combine multiple arrays into one, without necessarily changing the underlying data or (in the reshape case) copying it.

reshape() — Change Shape, Same Data

arr = np.arange(12)          # [0, 1, ..., 11], shape (12,)
arr.reshape(3, 4)              # shape (3, 4)
arr.reshape(4, 3)                # shape (4, 3)
arr.reshape(2, -1)                 # -1 means "infer this dimension" -> (2, 6)
arr.reshape(-1, 1)                   # column vector, shape (12, 1)

reshape returns a view when possible

If the data is contiguous in memory, reshape() doesn’t copy — it just reinterprets the shape. Modifying the reshaped array can modify the original. Use .reshape(...).copy() if independence is needed.

flatten() vs ravel()

arr2d.flatten()      # always returns a COPY, 1D
arr2d.ravel()          # returns a VIEW when possible (faster, but may mutate original)

transpose() / .T

arr2d.T                     # swap axes (rows <-> columns)
arr2d.transpose()             # equivalent
arr3d.transpose(2, 0, 1)        # explicit axis reordering for ND arrays
np.swapaxes(arr3d, 0, 2)          # swap two specific axes

Adding/Removing Dimensions

arr = np.array([1, 2, 3])          # shape (3,)
arr[:, np.newaxis]                    # shape (3, 1)
np.expand_dims(arr, axis=0)              # shape (1, 3)
np.expand_dims(arr, axis=1)                # shape (3, 1)
 
arr2d = np.array([[1, 2, 3]])          # shape (1, 3)
np.squeeze(arr2d)                          # removes size-1 dims -> shape (3,)

Concatenation & Stacking

a = np.array([1, 2, 3])
b = np.array([4, 5, 6])
 
np.concatenate([a, b])               # [1,2,3,4,5,6] — join along existing axis
np.concatenate([arr2d_a, arr2d_b], axis=0)   # stack rows (vertical)
np.concatenate([arr2d_a, arr2d_b], axis=1)     # stack columns (horizontal)
 
np.vstack([a, b])          # stack as rows -> shape (2, 3)
np.hstack([a, b])            # stack side by side -> shape (6,)
np.column_stack([a, b])        # stack as columns -> shape (3, 2)
np.stack([a, b], axis=0)         # NEW axis created -> shape (2, 3)
FunctionBehavior
concatenatejoins along an existing axis
stackjoins along a new axis (increases dimensionality)
vstackshorthand for row-wise concatenation
hstackshorthand for column-wise concatenation

Splitting

np.split(arr, 3)                    # split into 3 equal parts
np.split(arr, [3, 7])                 # split at specific indices -> 3 unequal parts
np.hsplit(arr2d, 2)                     # split horizontally (by columns)
np.vsplit(arr2d, 2)                       # split vertically (by rows)

Repeating & Tiling

np.repeat(arr, 3)             # repeat each element 3 times: [1,1,1,2,2,2,3,3,3]
np.tile(arr, 3)                 # repeat the whole array 3 times: [1,2,3,1,2,3,1,2,3]
np.tile(arr2d, (2, 3))            # tile a 2D array 2x vertically, 3x horizontally

Resizing (Changes Total Element Count)

np.resize(arr, (3, 4))    # reshape, repeating/truncating data as needed to fit new size
arr.resize((3, 4))           # in-place version (fills with 0 if growing)

resize vs reshape

reshape() requires the new shape to have the same total element count; resize() allows changing the total count, padding with zeros or repeating/truncating data — a common source of confusion.

Graph View

Backlinks