Directory

basic
研一机器学习相关课程作业汇总

1.感知器算法
2.多元线性回归
3.贝叶斯垃圾短信过滤
4.新闻推荐(未完成)
5.Django垃圾短信分类
6.爬虫

Machine Learning
机器学习基石&技法 课程笔记和作业
When can Machines Learn?

1.机器学习问题
2.二分类
3.不同的ML类型
4.可行性
hw0: 条件概率 and 贝叶斯公式
hw1: Perceptron and Pocket algorithm实现

Why can Machines Learn?

1.growthFunction,breakPoint
2.ML泛化理论
3.VC维度,边界
4.噪声和错误估计
hw2:错误率 VC bound计算样本数目N

How can Machines Learn?

1.线性回归,伪逆矩阵,squaredError
2.逻辑回归,sidmod函数,crossEntropyError
3.多分类问题,SGD
4.非线性问题的featureTransform
hw3:损失函数,linear/logistic(SGD) algorithm实现

How can Machines Learn Better?

1.过度拟合的危害,避免的方法
2.有约束的regularizer
3.验证集validation作用
4.小技巧,课程总结
hw4:添加项regularization,验证集valiadation的实现,计算

How can machines learn by Embedding numerous features

1.线性SVM,推导非条件目标,QP求解
2.对偶SVM,非线性问题消除Z域d+1依赖
3.kernel trick仅在X域计算
4.soft-margin,ξn
5.KLR,two-level-learning模拟Z域逻辑回归
6.SVR,tube regression
hw1:soft-margin SVM分类,linear,poly,rbf实验

How can machines learn by Combining predictive features

1.blending,bagging,bootstrap获取多样性gt
2.adaboost,惩罚因子Ut
3.decisionTree,impurity衡量
4.randomForest,feature-selection
5.GBDT,residual fitting
hw2:Adaboost-stump 未完成 hw3:cart tree,random forest 未完成

how can machines learn by distilling hidden features?

1.NeuralNetwork,backprop,optimization Tricks
2.DeepLearning,pre-trained autoencoder,denoising
3.RBFnetwork,distance similarity,k-means algorithm
4.linear network,alternating leastSQR
5.feature exaction,optimization,overfitting
hw4:NNet 未完成,k-nearest-means,k-means

Deep Learning
CS231n课程笔记和作业
https://github.com/cuixuage/2018SpringCS231n

Reinforcement Learning
RLAI 书籍阅读和代码实现
https://github.com/cuixuage/Reinforcement_Learning

Recommend

1.GBDT (xgboost实现)
2.FM (tensorflow实现)
3.FFM
4.diversity(MMR & DPP 实现)

Rreference

1.机器学习基石&技法 Prof. Hsuan-Tien Lin
2.深度学习CS231n Prof. Fei-Fei Li
3.强化学习RLAI Prof. Richard S.Sutton
4.推荐算法入门级代码 知乎——点击率预估代码
5.计算广告论文 论文汇总