Reshaping and Joining Arrays

Data rarely arrives in the shape you need. A model may want a column instead of a row, or you may need to glue two months of data together. This lesson covers the shape tools.

reshape: same numbers, new shape#

reshape(rows, cols) rearranges the elements without changing them. The total must stay the same: 12 elements can be 3 × 4, 4 × 3, 2 × 6 or 6 × 2, but not 5 × 2.

import numpy as np

a = np.arange(1, 13)
print(a.reshape(3, 4))
print(a.reshape(2, -1))

Output:

[[ 1  2  3  4]
 [ 5  6  7  8]
 [ 9 10 11 12]]
[[ 1  2  3  4  5  6]
 [ 7  8  9 10 11 12]]

-1 means “work it out for me”: with 12 elements and 2 rows, NumPy computes 6 columns.

A very common case in machine learning: turning a 1-D list of values into one column, because scikit-learn expects a table of features:

import numpy as np

sizes = np.array([650, 800, 1200])
X = sizes.reshape(-1, 1)
print(X)
print(X.shape)

Output:

[[ 650]
 [ 800]
 [1200]]
(3, 1)

Flatten and transpose#

import numpy as np

m = np.array([[1, 2, 3],
              [4, 5, 6]])
print(m.flatten())
print(m.T)
print(m.T.shape)

Output:

[1 2 3 4 5 6]
[[1 4]
 [2 5]
 [3 6]]
(3, 2)

• flatten() turns any shape back into one long row.
• .T (transpose) swaps rows and columns: row 0 becomes column 0.

Joining arrays#

import numpy as np

jan = np.array([[10, 12], [14, 9]])
feb = np.array([[11, 15], [13, 8]])

print(np.vstack([jan, feb]))
print(np.hstack([jan, feb]))

Output:

[[10 12]
 [14  9]
 [11 15]
 [13  8]]
[[10 12 11 15]
 [14  9 13  8]]

• vstack (vertical) puts arrays on top of each other: more rows.
• hstack (horizontal) puts them side by side: more columns.
• np.concatenate([a, b], axis=0) is the general version: axis=0 stacks rows, axis=1 stacks columns.

Remember

axis=0 goes down the rows, axis=1 goes across the columns. You’ll meet axis again with sums and averages.

Splitting#

import numpy as np

a = np.arange(10)
first, second = np.split(a, 2)
print(first, second)

Output:

[0 1 2 3 4] [5 6 7 8 9]

Practice#

1. Reshape np.arange(24) into 4 rows and let NumPy find the columns.
2. Stack the arrays [1, 2, 3] and [4, 5, 6] into a 2 × 3 table.
3. What is the shape of the transpose of a (5, 2) array?

Summary#

Task	Code
New shape, same data	a.reshape(r, c) (use -1 for one size)
One long row	a.flatten()
Swap rows and columns	a.T
Stack as rows / columns	np.vstack, np.hstack
Split into parts	np.split(a, n)