TopolearnData rarely arrives in the shape an operation needs; reshaping is how you line it up without copying.
An array's shape is just a view of a flat sequence of values. a.reshape(new_shape) reinterprets the same elements with new dimensions, as long as the total size matches. One dimension may be -1, meaning "work it out": a.reshape(3, -1).
Order matters. By default NumPy fills and reads in row-major (C) order: the last axis changes fastest. np.arange(6).reshape(2, 3) gives rows [0 1 2] and [3 4 5].
Related tools:
a.ravel()flattens to 1-D, returning a view when possible;a.flatten()always copies.a.T(ornp.transpose) reverses the axes; for 2-D that swaps rows and columns.a[:, np.newaxis]ora[:, None]inserts a length-1 axis, turning a shape(3,)vector into a(3, 1)column.np.expand_dimsandnp.squeezeadd and remove such axes by position.
Reshape and transpose usually return views: no data is copied, and writing through the view changes the original. That makes them nearly free, even on huge arrays.
A common confusion: a 1-D array of shape (3,) is neither a row nor a column. Transposing it does nothing. If an operation needs an explicit column, reshape to (3, 1).
Axes are numbered from 0. For a shape (2, 3) array, axis 0 runs down the rows (length 2) and axis 1 across the columns (length 3). That numbering is used everywhere: in aggregations, concatenation and broadcasting.
Example
import numpy as np
a = np.arange(6)
m = a.reshape(2, 3)
print(m)
# [[0 1 2]
# [3 4 5]]
print(a.reshape(3, -1).shape) # (3, 2)
print(m.T.shape) # (3, 2)
print(m.ravel()) # [0 1 2 3 4 5]
v = np.array([1, 2, 3])
print(v.T.shape) # (3,) -- 1-D: transpose is a no-op
col = v[:, np.newaxis]
print(col.shape) # (3, 1)
print(np.squeeze(col).shape) # (3,)
m[0, 0] = 99 # m is a view of a
print(a[0]) # 99Remember
reshapekeeps the data and changes the shape; use -1 for one inferred axis- Default ordering is row-major: the last axis varies fastest
None/np.newaxisadds a length-1 axis;squeezeremoves them- Reshape, ravel and
.Tusually return views, not copies
Watch out for
- Calling
.Ton a 1-D array to make a column vector; reshape to(n, 1)instead.