看不懂的python矩阵处理

最新推荐文章于 2022-10-27 20:53:07 发布

qq_17136695

最新推荐文章于 2022-10-27 20:53:07 发布

阅读量377

点赞数

分类专栏：问题

本文链接：https://blog.csdn.net/qq_17136695/article/details/52901706

版权

问题专栏收录该内容

2 篇文章 0 订阅

订阅专栏

X_vec_squared = np.diag( np.dot(X, X.transpose()) )

X_train_vec_squared = np.diag( np.dot(self.X_train, self.X_train.transpose()) )

X_X_train = -2*np.dot(self.X_train, X.transpose())

dists = np.sqrt( (X_X_train + X_vec_squared).transpose() + X_train_vec_squared )

X is np with(500,1000) X_train is np with (5000,1000) 对于X中的每一个元素（一行），和X_train中每一个元素，进行norm2运算

最终得到（500,5000）dist矩阵。

eg:

>>> a=np.array([[1,2],[2,3],[3,3]])
>>> b=np.array([[3,2],[1,3],[4,3],[1,1]])
>>> a_vec_squared = np.diag( np.dot(a, a.transpose()) )
>>> a_vec_squared
array([ 5, 13, 18])
>>> b_train_vec_squared = np.diag( np.dot(b, b.transpose()) )
>>> b_train_vec_squared
array([13, 10, 25, 2])

>>> ab_train = -2*np.dot(b, a.transpose())
>>> ab_train
array([[-14, -24, -30],
[-14, -22, -24],
[-20, -34, -42],
[ -6, -10, -12]])
>>> dists = np.sqrt( (ab_train + a_vec_squared).transpose() + b_train_vec_squared )
>>> dists
array([[ 2. , 1. , 3.16227766, 1. ],
[ 1.41421356, 1. , 2. , 2.23606798],
[ 1. , 2. , 1. , 2.82842712]])