LEM（拉普拉斯特征映射）算法，拉普拉斯特征映射是基于局部邻域，保持局部结构的流形学习方法。LEM通过一个无向加权图刻画流形上数据点间的近邻关系，图的顶点为原始数据点，图的边对应点之间的近邻关系，边的权值对应近邻点之间的相似程度（也可以是某种距离），LEM在低维嵌入空间中尽量保持图中数据点之间的近邻关系，然后求取嵌入坐标。通俗的说，LEM认为在高维数据空间离得近的点在低维嵌入空间也应该离得近-LEM (Laplace feature mapping) algorithm, Laplace feature mapping is based on a local neighborhood, holding manifold learning method local structure. LEM through a non-weighted graph depicts flow to form a neighbor relationship between data points, the graph' s vertex to the original data points, corresponding points edge neighbor relationship graph between the edge weights corresponding to the degree of similarity between neighboring points ( may be some distance), LEM keep neighbor relationship diagram between the data points in a low dimensional embedding space, and then strike embedding coordinates. Popular to say, LEM believes in high-dimensional data space be closer point in the low-dimensional embedding space should be closer