一种使用numpy的热编码 [英] One Hot Encoding using numpy
问题描述
如果输入为零,我想创建一个看起来像这样的数组:
If the input is zero I want to make an array which looks like this:
[1,0,0,0,0,0,0,0,0,0]
,如果输入为5:
[0,0,0,0,0,1,0,0,0,0]
对于上述内容,我写道:
For the above I wrote:
np.put(np.zeros(10),5,1)
但是没有用.
有什么方法可以在一排中实现?
Is there any way in which, this can be implemented in one line?
推荐答案
通常,当您希望在机器学习中获得一种用于分类的单编码时,您将拥有一个索引数组.
Usually, when you want to get a one-hot encoding for classification in machine learning, you have an array of indices.
import numpy as np
nb_classes = 6
targets = np.array([[2, 3, 4, 0]]).reshape(-1)
one_hot_targets = np.eye(nb_classes)[targets]
one_hot_targets
现在是
array([[[ 0., 0., 1., 0., 0., 0.],
[ 0., 0., 0., 1., 0., 0.],
[ 0., 0., 0., 0., 1., 0.],
[ 1., 0., 0., 0., 0., 0.]]])
.reshape(-1)
可以确保使用正确的标签格式(您可能还具有[[2], [3], [4], [0]]
). -1
是一个特殊值,表示将所有剩余的物料放入此维".因为只有一个,所以它使数组变平.
The .reshape(-1)
is there to make sure you have the right labels format (you might also have [[2], [3], [4], [0]]
). The -1
is a special value which means "put all remaining stuff in this dimension". As there is only one, it flattens the array.
def get_one_hot(targets, nb_classes):
res = np.eye(nb_classes)[np.array(targets).reshape(-1)]
return res.reshape(list(targets.shape)+[nb_classes])
包裹
您可以使用 mpu.ml.indices2one_hot .经过测试且易于使用:
Package
You can use mpu.ml.indices2one_hot. It's tested and simple to use:
import mpu.ml
one_hot = mpu.ml.indices2one_hot([1, 3, 0], nb_classes=5)
这篇关于一种使用numpy的热编码的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!