[科研] | 101级别解释 | One-hot

转载已于 2023-03-06 18:10:06 修改 · 192 阅读

CC 4.0 BY-SA版权

原文链接：https://en.wikipedia.org/wiki/One-hot

文章标签：

于 2023-03-02 17:53:04 首次发布

One-Hot编码是一种在自然语言处理中用于区分词汇的技术，它创建了一个1×N矩阵，其中每个词由一个独特的1表示，其余位置为0。这种方法确保了机器学习模型不会错误地认为数字大小有特定的重要性。

直觉解释

A one-hot is a group of bits among which the legal combinations of values are only those with a single high(1) bit and all the others low(0).

A one-hot vector is a $1×N1\times N$ matrix (vector) used to distinguish each word in a vocabulary from every other word in the vocabulary. The vector consists of 0s in all cells with the exception of a single 1 in a cell used uniquely to identify the word. One-hot encoding ensures that machine learning does not assume that higher numbers are more important.