高效地将概率列表转换为0/1列表，只需取最高概率的一部分%而无需重新索引

huangapple
go评论90阅读模式
英文:

Efficiently convert list of probabilities in a list of 0/1 by taking a % of highest probabilities without reindexing

问题

Sure, here are the translated code parts you requested:

  1. # Problem
  2. 给定一个大型的概率数组和要提取的百分比
  3. ```python3
  4. probabilities = [0.1, 0.4, 0.7, 0.2, 0.9, 0.5, 0.6]
  5. N_percentages = [20, 30, 40]  # 列表大小的百分比

我想要高效地计算

  1. {20:[0, 0, 0, 0, 1, 0, 0], 30:[0, 0, 1, 0, 1, 0, 0], 40:[0, 0, 1, 0, 1, 0, 1]}

我不能失去索引 - 值必须保持原来的位置

我的尝试到目前为止：

解决方案1

  1. def mark_probabilities_for_multiple_N1(probabilities, N_percentages):
  2.     marked_lists = {}
  3.     sorted_indices = sorted(range(len(probabilities)), key=lambda i: probabilities[i], reverse=True)
  4.     list_size = len(probabilities)
  5.     
  6.     for N_percentage in N_percentages:
  7.         N = int(N_percentage * list_size / 100)
  8.         
  9.         marked_lists[N_percentage] = [1 if i in sorted_indices[:N] else 0 for i in range(len(probabilities))]
  11.         # 利用先前计算的较小N值的标记列表
  12.         for prev_N_percentage in [prev_N for prev_N in marked_lists if prev_N < N_percentage]:
  13.             marked_lists[N_percentage] = [1 if marked_lists[prev_N_percentage][i] == 1 or marked_lists[N_percentage][i] == 1 else 0 for i in range(len(probabilities))]
  15.     return marked_lists

解决方案2 - 使用heapq

将（索引，概率值）映射到heapq，按probability_value排序

  1. def indicies_n_largest(values_with_indicies, percentage) -> set[int]:  # O(1) exists(int)
  2.     """
  3.     返回数组中n个最大概率的索引列表。
  5.     :param arr: 概率数组
  6.     :param percentage: 要返回的最大概率的百分比
  7.     返回：最大概率的索引列表
  8.     """
  9.     fraction = percentage / 100
  10.     samples_num = int(len(values_with_indicies) * fraction)
  11.     result = heapq.nlargest(samples_num, values_with_indicies, key=lambda x: x[1])
  12.     return [x[0] for x in result]
  15. def percentage_indicies_map(action_probs, percentages) -> dict[int, set[int]]:
  16.     """
  17.     给定动作概率和百分比列表，返回一个动作索引的映射，这些动作被视为优秀，对于每个百分比。
  18.     """
  19.     values_wth_indicies = [(i, x) for i, x in enumerate(action_probs)]
  21.     percentage_indicies_map: dict[
  22.         int, set[int]
  23.     ] = {}  # 最大概率的索引列表
  25.     for percentage in percentages:
  26.         percentage_indicies_map[percentage] = indicies_n_largest(values_wth_indicies, percentage)
  28.     return percentage_indicies_map

请注意，我只提供了代码的翻译部分，没有包括问题或其他内容。如果您需要进一步的说明或帮助，请告诉我。

英文:

Problem

Given a huge array of probabilities and the percentages to take

  1. probabilities = [0.1, 0.4, 0.7, 0.2, 0.9, 0.5, 0.6]
  2. N_percentages = [20, 30, 40]  # Percentage of the list size

I want to efficiently compute

  1. {20:[0, 0, 0, 0, 1, 0, 0], 30:[0, 0, 1, 0, 1, 0, 0], 40:[0, 0, 1, 0, 1, 0, 1]}

I cannot lose the indexing - values have to keep their original place

My tries so far:

Solution number 1

  1. def mark_probabilities_for_multiple_N1(probabilities, N_percentages):
  2.     marked_lists = {}
  3.     sorted_indices = sorted(range(len(probabilities)), key=lambda i: probabilities[i], reverse=True)
  4.     list_size = len(probabilities)
  5.     
  6.     for N_percentage in N_percentages:
  7.         N = int(N_percentage * list_size / 100)
  8.         
  9.         marked_lists[N_percentage] = [1 if i in sorted_indices[:N] else 0 for i in range(len(probabilities))]
  10.         
  11.         # Utilize previously calculated marked lists for smaller N values
  12.         for prev_N_percentage in [prev_N for prev_N in marked_lists if prev_N &lt; N_percentage]:
  13.             marked_lists[N_percentage] = [1 if marked_lists[prev_N_percentage][i] == 1 or marked_lists[N_percentage][i] == 1 else 0 for i in range(len(probabilities))]
  15.     return marked_lists

Solution number 2 - use heapq

Map the (idx, probability_value) to a heapq, order by the probability_value

  1. def indicies_n_largest(values_with_indicies, percentage) -&gt; set[int]:  # O(1) exists(int)
  2.     &quot;&quot;&quot;
  3.     Returns a list of indicies for n largest probabilities in the array.
  5.     :param arr: array of probabilities
  6.     :param percentage: percentage of the largest probabilities to be returned
  7.     returns: list of indicies of the largest probabilities
  8.     &quot;&quot;&quot;
  9.     fraction = percentage / 100
  10.     samples_num = int(len(values_with_indicies) * fraction)
  11.     result = heapq.nlargest(samples_num, values_with_indicies, key=lambda x: x[1])
  12.     return [x[0] for x in result]
  15. def percentage_indicies_map(action_probs, percentages) -&gt; dict[int, set[int]]:
  16.     &quot;&quot;&quot;
  17.     Given action probabilities and a list of percentages, return a map of actions&#39; indicies that are considered good,
  18.     for each percentage.
  19.     &quot;&quot;&quot;
  20.     values_wth_indicies = [(i, x) for i, x in enumerate(action_probs)]
  22.     percentage_indicies_map: dict[
  23.         int, set[int]
  24.     ] = {}  # list of indicies of the largest probabilities
  26.     for percentage in percentages:
  27.         percentage_indicies_map[percentage] = indicies_n_largest(values_wth_indicies, percentage)
  29.     return percentage_indicies_map

答案1

得分: 1

以下是您要翻译的代码部分：

  1. probabilities = [0.1, 0.4, 0.7, 0.2, 0.9, 0.5, 0.6]
  2. N_percentages = [20, 30, 40]
  4. out, s = {}, sorted(enumerate(probabilities), key=lambda k: -k[1])
  5. for p in N_percentages:
  6. 	ones = set(i for i, _ in s[:round((p / 100) * len(probabilities))])
  7. 	out
     = [int(i in ones) for i in range(len(probabilities))]
  9. print(out)

打印输出：

  1. {
  2.   20: [0, 0, 0, 0, 1, 0, 0], 
  3.   30: [0, 0, 1, 0, 1, 0, 0], 
  4.   40: [0, 0, 1, 0, 1, 0, 1]
  5. }
英文:

You can try:

  1. probabilities = [0.1, 0.4, 0.7, 0.2, 0.9, 0.5, 0.6]
  2. N_percentages = [20, 30, 40]
  4. out, s = {}, sorted(enumerate(probabilities), key=lambda k: -k[1])
  5. for p in N_percentages:
  6. 	ones = set(i for i, _ in s[:round((p / 100) * len(probabilities))])
  7. 	out
     = [int(i in ones) for i in range(len(probabilities))]
  9. print(out)

Prints:

  1. {
  2.   20: [0, 0, 0, 0, 1, 0, 0], 
  3.   30: [0, 0, 1, 0, 1, 0, 0], 
  4.   40: [0, 0, 1, 0, 1, 0, 1]
  5. }

huangapple
  • 本文由 发表于 2023年5月21日 06:58:37
  • 转载请务必保留本文链接：https://go.coder-hub.com/76297638.html
  • classification
  • data-structures
  • heapq
  • python-3.x
  • sorting
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定