嵌套字典中包含的向量的相关性

Correlation of vectors contained in nested dictionaries

我有一个具有下一个结构的嵌套字典:

{Cell_name_1 : {KPI_name_1: [value1, value2, ..., valueN], 
                KPI_name_2: [value1, value2, ..., valueN], 
                ..., 
                KPI_name_N: [value1, value2, ..., valueN]}, 
 Cell_name_2 : {KPI_name_1: [value1, value2, ..., valueN], ...}, 
 Cell_name_N : {....}}

我想检查不同单元格中包含的 vectos 之间的相关性(我已经定义了这个方法,所以它是一个辅助函数)。比方说:

vector_1 = [64.0, 66.0, 53.5, 52.1, 54.0] #[values from KPI_name_1 from Cell_name_1]
vector_2 = [84.0, 86.0, 63.5, 72.1, 24.0] #[values from KPI_name_2 from Cell_name_2]

correlation(vector_1, vector_2)

我尝试了不同的字典循环方式(普通 for 循环、带 while 和条件的经典循环等),但我没有找到我需要的方法。

举个例子,代码是这样的:

dic_sem = {'16895555': {'KPI_name_1': [64.0, 66.0, 53.5, 52.1, 54.0], 
                        'KPI_name_2': [54.0, 56.0, 23.5, 32.1, 84.0]}, 
           '16894444': {'KPI_name_1': [84.0, 86.0, 63.5, 72.1, 24.0], 
                        'KPI_name_2': [24.0, 26.0, 63.5, 92.1, 84.0]}}

'16895555''16894444'是不同的Cell_name's

您可以遍历字典并创建单元名称字典,例如KPI_name_1 到包含您的向量的列表列表

from collections import defaultdict

vectors = defaultdict(list)

#Iterate over the values
for value in dic_sem.values():
    #Create your vectors dictionary
    for k, v in value.items():
        vectors[k].append(v)

print(dict(vectors))

输出将是

{'KPI_name_1': [[64.0, 66.0, 53.5, 52.1, 54.0], [84.0, 86.0, 63.5, 72.1, 24.0]], 
'KPI_name_2': [[54.0, 56.0, 23.5, 32.1, 84.0], [24.0, 26.0, 63.5, 92.1, 84.0]]}

然后您可以迭代此字典的值并相应地调用 correlation

for value in vectors.values():
    print(value[0], value[1])
    #correlation(*value)

这里的输出将是

[64.0, 66.0, 53.5, 52.1, 54.0] [84.0, 86.0, 63.5, 72.1, 24.0]
[54.0, 56.0, 23.5, 32.1, 84.0] [24.0, 26.0, 63.5, 92.1, 84.0]

也许 itertools.product 可以帮到您:


  import itertools
  import numpy as np

  # Get vector names (assuming keys present in all cells)
  field_names = list(dic_sem.values())[0].keys()

  # Precompute all pairs of cells 
  all_cell_pairs = list(itertools.product(dic_sem.keys(), dic_sem.keys()))

  corr = {}
  for field in field_names: 
      corr[field] = np.reshape([correlation(dic_sem[c1][field], dic_sem[c2][field]) for c1, c2 in all_cell_pairs], (len(dic_sem), -1))

请注意,我们在这里进行了两倍以上的必要计算:相关矩阵是对称的,因此足以仅计算上三角或下三角(例如使用 itertools.combinations),不包括对角线(等于 1)。不过上面应该给了方向..