列表/数组字符串到numpy浮点数组

Question

我是scikit学习和numpy的新手。我怎么能代表我的数据集由列表/字符串数组组成，例如

[["aa bb","a","bbb","à"], [bb cc","c","ddd","à"], ["kkk","a","","a"]] 

给一个numpy数组的dtype float？

Answer 1

我认为你所寻找的是你的单词的数字表示。您可以使用gensim并将每个单词映射到令牌id，然后从中创建您的numpy阵列，如下所示：

import numpy as np 
from gensim import corpora 

toconvert = [["aa bb","a","bbb","à"], ["bb", "cc","c","ddd","à"], ["kkk","a","","a"]] 

# convert your list of lists into token id's. For example, 'aa bb' could be represented as a 2, a as a 1, etc. 
tdict = corpora.Dictionary(toconvert) 

# given nested structure, you can append nested numpy arrays 
newlist = [] 
for l in toconvert: 
    tmplist = [] 
    for word in l: 
     # append to intermediate list the id for the given word under observation 
     tmplist.append(tdict.token2id[word]) 
    # convert to numpy array and append to main list 
    newlist.append(np.array(tmplist).astype(float)) # type float 

print(newlist) # desired output: [array([ 2., 0., 1., 0.]), array([ 5., 3., 4., 6., 0.]), array([ 7., 0., 8., 0.])] 

# and to see what id's represent which strings: 
tdict[0] # 'a'