python - 如何从 numpy 矩阵中随机选择项目(以矢量化方式)？

Question

我有 numpy 的产品票价矩阵，比如

|--------|---------------------|------------------|------------------|------------------|
|        |     Product 1       |     Product 2    |     Product 3    |     Product 4    |
|--------|---------------------|------------------|------------------|------------------|
| Fare 1 |          10         |         11       |         12       |         13       |
|--------|---------------------|------------------|------------------|------------------|
| Fare 2 |          20         |         21       |         22       |         23       |
|--------|---------------------|------------------|------------------|------------------|
| Fare 3 |          30         |         31       |         32       |         33       |
|--------|---------------------|------------------|------------------|------------------|

我的目标是在连续几天内为每种产品随机选择票价。本质上，我希望有类似的东西

|-------|---------------------|------------------|------------------|------------------|
|       |     Product 1       |     Product 2    |     Product 3    |     Product 4    |
|-------|---------------------|------------------|------------------|------------------|
| Day 1 |          30         |         11       |         22       |         13       |
|-------|---------------------|------------------|------------------|------------------|
| Day 2 |          30         |         31       |         22       |         33       |
|-------|---------------------|------------------|------------------|------------------|

我想到的唯一方法是使用经典循环:

import numpy as np

generator = np.random.default_rng()
days = {1, 2}
fare_product_matrix = np.array([[10, 20, 30], [11, 21 , 31], [12, 22, 32], [13, 23, 33]])

fare_indexes = generator.integers(0, fare_product_matrix.shape[1], size=(len(days), fare_product_matrix.shape[0]))

out = np.empty(fare_indexes.shape)
for day in range(len(days)):
    for product in range(fare_product_matrix.shape[0]):
        out[day, product] = fare_product_matrix[product, fare_indexes[day, product]]

是否有任何智能矢量化方法可以用 numpy 做到这一点？

Answer 1

像这样的东西:

rows = np.random.randint(0,fare_product_matrix.shape[0], 
                        len(days) * fare_product_matrix.shape[1])

cols = np.tile(np.arange(fare_product_matrix.shape[1]), len(days))

fare_product_matrix[rows, cols].reshape(len(days), -1)

输出:

array([[30, 21, 12, 13],
       [10, 31, 32, 23]])