使用带有广播或各种数组形状的 numpy select

我有各种数组形状，我尝试使用 np.select()。我的代码 selected_results 适用于这种情况：

# Given arrays
results = [[array(['alpha'], dtype=object) array([0.16134485])
  array([-2.19302435,  0.57976273]) array([1.25348942, 0.77778261])
  array([0.56136549, 0.56136549]) array([1.12076068, 0.53834429])]
 [array(['alpha'], dtype=object) array([0.11778484])
  array([-0.86983912,  0.24081955]) array([0.84209155, 0.84209155])
  array([0.66394431, 0.66394431]) array([0.51797309, 0.51797309])]
 [array(['alpha'], dtype=object) array([0.11920295])
  array([-2.5218011 ,  0.25602172]) array([1.02998263, 0.88191816])
  array([0.63414049, 0.63414049]) array([0.81162187, 0.61289923])]]

conditions = [array([False, False]), array([False, False]), array([ True,  True])]

choices = [array([array(['alpha'], dtype=object), array(['alpha'], dtype=object),
       array(['alpha'], dtype=object)], dtype=object), array([array([0.16134485]), array([0.11778484]), array([0.11920295])],
      dtype=object), array([array([-2.19302435,  0.57976273]),
       array([-0.86983912,  0.24081955]),
       array([-2.5218011 ,  0.25602172])], dtype=object), array([array([1.25348942, 0.77778261]), array([0.84209155, 0.84209155]),
       array([1.02998263, 0.88191816])], dtype=object), array([array([0.56136549, 0.56136549]), array([0.66394431, 0.66394431]),
       array([0.63414049, 0.63414049])], dtype=object), array([array([1.12076068, 0.53834429]), array([0.51797309, 0.51797309]),
       array([0.81162187, 0.61289923])], dtype=object)]

selected_results = np.array([np.select(conditions, choice, default=np.nan) for choice in choices])
print(selected_results)

>> [['alpha' 'alpha']
 [0.119202946 0.119202946]
 [-2.521801101837287 0.25602172377257837]
 [1.0299826279625939 0.881918155681708]
 [0.63414049 0.63414049]
 [0.8116218656784026 0.612899235]]

但是，当我有：

# Given arrays
results = [['alpha']
 [0.161344848]
 [-2.19302435146961]
 [1.2534894161924872]
 [0.5613654925053716]
 [1.120760679329349]
 ['alpha']
 [0.117784836]
 [-0.8698391195603491]
 [0.8420915464087234]
 [0.663944309731699]
 [0.517973094]
 ['alpha']
 [0.119202946]
 [-2.521801101837287]
 [1.0299826279625939]
 [0.63414049]
 [0.8116218656784026]]

conditions = [array([False]), array([False]), array([ True])]

choices = [array(['alpha', 0.161344848, -2.19302435146961, 1.2534894161924872,
       0.5613654925053716, 1.120760679329349, 'alpha', 0.117784836,
       -0.8698391195603491, 0.8420915464087234, 0.663944309731699,
       0.517973094, 'alpha', 0.119202946, -2.521801101837287,
       1.0299826279625939, 0.63414049, 0.8116218656784026], dtype=object)]

使用 selected_results = np.array([np.select(conditions, choice, default=np.nan) for choice in choices]) 给出 ValueError: list of cases must be same length as list of conditions

预期的输出是：

>> [['alpha']
 [0.119202946]
 [-2.521801101837287]
 [1.0299826279625939]
 [0.63414049]
 [0.8116218656784026]]

我知道有一个问题形状，但我不知道如何为“selected_results”提供一个更普遍适用的简单的一两行解决方案。

问题在于，在你的第二种情况下， conditions 中有 3 个条件，但你的 choices 数组中每个元素只有 1 个选择。为了解决这个问题，需要为 choices 中的每个条件提供一个选择。

你可以通过将你的单个 choice 数组分成多个大小相等的块来实现。由于你有 3 个条件，因此你应该将 choice 数组分成 3 个相等的块，每个块代表一个条件的选择。

以下是如何做到这一点：

import numpy as np

results = ['alpha', 0.161344848, -2.19302435146961, 1.2534894161924872, 0.5613654925053716, 1.120760679329349, 
           'alpha', 0.117784836, -0.8698391195603491, 0.8420915464087234, 0.663944309731699, 0.517973094, 
           'alpha', 0.119202946, -2.521801101837287, 1.0299826279625939, 0.63414049, 0.8116218656784026]

conditions = np.array([False, False, True])

# 将 choices 数组分成 3 个相等的块
choices = np.array_split(results, len(conditions))

# 使用 np.select 选择每个块中相应条件的值
selected_results = np.array([np.select([conditions[i]], [choice], default=np.nan) for i, choice in enumerate(choices)])

# 将结果展平为一维数组
selected_results = selected_results.flatten()

print(selected_results)

这将输出预期的结果：

['alpha' 0.119202946 -2.521801101837287 1.0299826279625939
 0.63414049 0.8116218656784026]

这个解决方案通过将 choices 数组拆分为每个条件对应一个块来解决数组形状问题。然后，它使用 np.select 为每个条件选择相应的块，并生成预期结果的一维数组。