我正在寻找一个由窗口大小为 N 的单词组成的字符串滑动窗口拆分器。
输入:"我喜欢食物,我喜欢喝酒",窗口尺寸 3
输出: [ "我爱食物","爱食物和","食物和我","我喜欢" .....]
窗口滑动的所有建议都是围绕字符串序列,没有术语。有什么开箱即用的东西吗?
您可以使用具有不同偏移量的迭代器并压缩所有偏移量。
>>> arr = "I love food. blah blah".split()
>>> its = [iter(arr), iter(arr[1:]), iter(arr[2:])] #Construct the pattern for longer windowss
>>> zip(*its)
[('I', 'love', 'food.'), ('love', 'food.', 'blah'), ('food.', 'blah', 'blah')]
如果你有很长的句子,或者可能是普通的旧循环(如另一个答案),你可能想使用izip
。
一种基于下标字符串序列的方法:
def split_on_window(sequence="I love food and I like drink", limit=4):
results = []
split_sequence = sequence.split()
iteration_length = len(split_sequence) - (limit - 1)
max_window_indicies = range(iteration_length)
for index in max_window_indicies:
results.append(split_sequence[index:index + limit])
return results
示例输出:
>>> split_on_window("I love food and I like drink", 3)
['I', 'love', 'food']
['love', 'food', 'and']
['food', 'and', 'I']
['and', 'I', 'like']
['I', 'like', 'drink']
以下是受@SuperSaiyan启发的替代答案:
from itertools import izip
def split_on_window(sequence, limit):
split_sequence = sequence.split()
iterators = [iter(split_sequence[index:]) for index in range(limit)]
return izip(*iterators)
示例输出:
>>> list(split_on_window(s, 4))
[('I', 'love', 'food', 'and'), ('love', 'food', 'and', 'I'),
('food', 'and', 'I', 'like'), ('and', 'I', 'like', 'drink')]
基准:
Sequence = I love food and I like drink, limit = 3
Repetitions = 1000000
Using subscripting -> 3.8326420784
Using izip -> 5.41380286217 # Modified to return a list for the benchmark.
def token_sliding_window(str, size):
tokens = str.split(' ')
for i in range(len(tokens )- size + 1):
yield tokens[i: i+size]