Broadcasting con NumPy
El broadcasting es una forma de hacer operaciones entre arrays de diferentes tamaños, que usualmente no son compatibles. El broadcasting es posible cuando se cumplen ciertas reglas.
- El tamaño de cada dimensión es igual.
- Una de las dimensiones es 1.
El broadcasting se realiza en la dimensión que tiene tamaño 1. El array con tamaño 1 se extiende para que tenga el mismo tamaño que el otro array.
Ejemplos de broadcasting
| Array 1 | Array 2 | Resultado |
|---|---|---|
| 10 x 5 | 1 x 5 | 10 x 5 |
| 10 x 5 | 5 | 10 x 5 |
| 10 x 5 | 10 x 1 | 10 x 5 |
| 10 x 5 | scalar | 10 x 5 |
import numpy as np
from rich import print
# matris de 10x5
m10_5 = np.random.randint(0, 10, (10, 5))
m1_5 = np.random.randint(0, 10, (1, 5))
m5 = np.random.randint(0, 10, (5))
m10_1 = np.random.randint(0, 10, (10, 1))
scalar = 5
print('Broadcasting de 10x5 + 1x5')
print(m10_5 + m1_5)
print('Broadcasting de 10x5 + 5')
print(m10_5 + m5)
print('Broadcasting de 10x5 + 10x1')
print(m10_5 + m10_1)
print(m10_5 + scalar)
Broadcasting de 10x5 + 1x5
[[ 9 2 6 5 1] [ 5 9 10 10 8] [ 6 9 3 10 3] [ 8 8 7 4 8] [ 5 2 6 10 5] [ 7 9 9 8 3] [ 7 5 3 5 7] [ 4 7 7 3 1] [ 2 8 6 12 7] [ 9 9 12 10 10]]
Broadcasting de 10x5 + 5
[[ 7 11 9 2 5] [ 3 18 13 7 12] [ 4 18 6 7 7] [ 6 17 10 1 12] [ 3 11 9 7 9] [ 5 18 12 5 7] [ 5 14 6 2 11] [ 2 16 10 0 5] [ 0 17 9 9 11] [ 7 18 15 7 14]]
Broadcasting de 10x5 + 10x1
[[14 9 10 9 7] [ 5 11 9 9 9] [ 5 10 1 8 3] [11 13 9 6 12] [ 9 8 9 13 10] [11 15 12 11 8] [ 6 6 1 3 7] [ 3 8 5 1 1] [ 8 16 11 17 14] [ 7 9 9 7 9]]
[[12 7 8 7 5] [ 8 14 12 12 12] [ 9 14 5 12 7] [11 13 9 6 12] [ 8 7 8 12 9] [10 14 11 10 7] [10 10 5 7 11] [ 7 12 9 5 5] [ 5 13 8 14 11] [12 14 14 12 14]]
Ejemplos de no broadcasting
| Array 1 | Array 2 | Resultado |
|---|---|---|
| 10 x 5 | 5 x 10 | Error |
| 10 x 5 | 10 | Error |
m10_5 = np.random.randint(0, 10, (10, 5))
m5_10 = np.random.randint(0, 10, (5, 10))
m10 = np.random.randint(0, 10, (10))
print('Broadcasting de 10x5 + 5x10')
print(m10_5 + m5_10)
Broadcasting de 10x5 + 5x10
ValueError: operands could not be broadcast together with shapes (10,5) (5,10)
print('Broadcasting de 10x5 + 10')
print(m10_5 + m10)
Broadcasting de 10x5 + 10
ValueError: operands could not be broadcast together with shapes (10,5) (10,)