海量数据挖掘MMDS week2: Nearest-Neighbor Learning最近邻学习

http://blog.csdn.net/pipisorry/article/details/48894963

海量数据挖掘Mining Massive Datasets(MMDs) -Jure Leskovec courses学习笔记之Nearest-Neighbor Learning，KNN最近邻学习

{The module is about large scale machine learning.}

Supervised Learning监督学习

Note: y有多种不同的形式，对应不同的问题。如为实数时，属于回归问题。

下面我们主要讲解分类问题

大规模机器学习方法

how do we efficiently train?Or build a model based on the based on the data?

So in a sense the main question is how do I find this function f.That takes the input features and predicts the class variable.

皮皮blog

Instance based learning基于实例的学习

最近邻分类器Nearest nerghbor

最近邻分类器要考虑的问题

Note: 最后一个要考虑的问题就是：How to take all these nearest neighbors and combine their values into a single point that I can use as prediction.

1-Nearest Nerghbor

1-Nearest nerghbor的重大缺陷：预测值附近变化大，用一个值来预测不准确。the method is suffering from It is making lots of very spiky, or sharp decisions, because we are only looking at the one nearest neighbor.

K-Nearest Nerghbor

Note: f(x) is much smoother than what is was before.

Kernel Regression核回归

皮皮blog

寻找最近邻的方法

一般扫描数据点方法的时间复杂度：线性时间

solution would require a linear pass over the data, so it would take linear time.

使用LSH的时间复杂度：常数时间（可用于大规模数据）

using locality sensitive hashing, we could find, nearest neighbors in near constant time.So that would be a good way how to really make nearest neighbor classifiers scale to large scale data.

具体是怎么实现的？

from:http://blog.csdn.net/pipisorry/article/details/48894963

ref:论文:GPU上的K近邻并行暴力搜索Brute-Force k-Nearest Neighbors Search on the GPU