【基础】常用的机器学习&数据挖掘知识点
转自http://www.36dsj.com/archives/20135
Basis(基础):
MSE(Mean Square Error 均方误差),LMS(LeastMean Square 最小均方),LSM(Least Square Methods 最小二乘法),MLE(MaximumLikelihood Estimation最大似然估计),QP(Quadratic Programming 二次规划), CP(Conditional Probability条件概率),JP(Joint Probability 联合概率),MP(Marginal Probability边缘概率),Bayesian Formula(贝叶斯公式),L1 /L2Regularization(L1/L2正则,以及更多的,现在比较火的L2.5正则等),GD(GradientDescent 梯度下降),SGD(Stochastic Gradient Descent 随机梯度下降),Eigenvalue(特征值),Eigenvector(特征向量),QR-decomposition(QR分解),Quantile (分位数),Covariance(协方差矩阵)。
Common Distribution(常见分布):
Discrete Distribution(离散型分布):BernoulliDistribution/Binomial(贝努利分布/二项分布),Negative BinomialDistribution(负二项分布),MultinomialDistribution(多项式分布),Geometric Distribution(几何分布),HypergeometricDistribution(超几何分布),Poisson Distribution (泊松分布)
Continuous Distribution (连续型分布):UniformDistribution(均匀分布),Normal Distribution /Guassian Distribution(正态分布/高斯分布),ExponentialDistribution(指数分布),Lognormal Distribution(对数正态分布),GammaDistribution(Gamma分布),Beta Distribution(Beta分布),Dirichlet Distribution(狄利克雷分布),Rayleigh Distribution(瑞利分布),Cauchy Distribution(柯西分布),Weibull Distribution (韦伯分布)
Three Sampling Distribution(三大抽样分布):Chi-squareDistribution(卡方分布),t-distribution(t-distribution),F-distribution(F-分布)
Data Pre-processing(数据预处理):
Missing Value Imputation(缺失值填充),Discretization(离散化),Mapping(映射),Normalization(归一化/标准化)。
Sampling(采样):
Simple Random Sampling(简单随机采样),OfflineSampling(离线等可能K采样),Online Sampling(在线等可能K采样),Ratio-based Sampling(等比例随机采样),Acceptance-RejectionSampling(接受-拒绝采样),Importance Sampling(重要性采样),MCMC(MarkovChain Monte Carlo 马尔科夫蒙特卡罗采样算法:Metropolis-Hasting& Gibbs)。
Clustering(聚类):
K-Means,K-Mediods,二分K-Means,FK-Means,Canopy,Spectral-KMeans(谱聚类),GMM-EM(混合高斯模型-期望最大化算法解决),K-Pototypes,CLARANS(基于划分),BIRCH(基于层次),CURE(基于层次),DBSCAN(基于密度),CLIQUE(基于密度和基于网格)
Classification&Regression(分类&回归):
LR(Linear Regression 线性回归),LR(LogisticRegression逻辑回归),SR(Softmax Regression 多分类逻辑回归),GLM(GeneralizedLinear Model 广义线性模型),RR(Ridge Regression 岭回归/L2正则最小二乘回归),LASSO(Least Absolute Shrinkage andSelectionator Operator L1正则最小二乘回归), RF(随机森林),DT(DecisionTree决策树),GBDT(Gradient BoostingDecision Tree 梯度下降决策树),CART(ClassificationAnd Regression Tree 分类回归树),KNN(K-Nearest Neighbor K近邻),SVM(Support VectorMachine),KF(KernelFunction 核函数PolynomialKernel Function 多项式核函数、Guassian KernelFunction 高斯核函数/Radial BasisFunction RBF径向基函数、String KernelFunction 字符串核函数)、 NB(Naive Bayes 朴素贝叶斯),BN(Bayesian Network/Bayesian Belief Network/ Belief Network 贝叶斯网络/贝叶斯信度网络/信念网络),LDA(Linear Discriminant Analysis/FisherLinear Discriminant 线性判别分析/Fisher线性判别),EL(Ensemble Learning集成学习Boosting,Bagging,Stacking),AdaBoost(Adaptive Boosting 自适应增强),MEM(MaximumEntropy Model最大熵模型)
Effectiveness Evaluation(分类效果评估):
Confusion Matrix(混淆矩阵),Precision(精确度),Recall(召回率),Accuracy(准确率),F-score(F得分),ROC Curve(ROC曲线),AUC(AUC面积),LiftCurve(Lift曲线) ,KS Curve(KS曲线)。
PGM(Probabilistic Graphical Models概率图模型):
BN(Bayesian Network/Bayesian Belief Network/ BeliefNetwork 贝叶斯网络/贝叶斯信度网络/信念网络),MC(Markov Chain 马尔科夫链),HMM(HiddenMarkov Model 马尔科夫模型),MEMM(Maximum Entropy Markov Model 最大熵马尔科夫模型),CRF(ConditionalRandom Field 条件随机场),MRF(MarkovRandom Field 马尔科夫随机场)。
NN(Neural Network神经网络):
ANN(Artificial Neural Network 人工神经网络),BP(Error BackPropagation 误差反向传播)
Deep Learning(深度学习):
Auto-encoder(自动编码器),SAE(Stacked Auto-encoders堆叠自动编码器:Sparse Auto-encoders稀疏自动编码器、Denoising Auto-encoders去噪自动编码器、Contractive Auto-encoders 收缩自动编码器),RBM(RestrictedBoltzmann Machine 受限玻尔兹曼机),DBN(Deep Belief Network 深度信念网络),CNN(ConvolutionalNeural Network 卷积神经网络),Word2Vec(词向量学习模型)。
DimensionalityReduction(降维):
LDA LinearDiscriminant Analysis/Fisher Linear Discriminant 线性判别分析/Fisher线性判别,PCA(Principal Component Analysis 主成分分析),ICA(IndependentComponent Analysis 独立成分分析),SVD(Singular Value Decomposition 奇异值分解),FA(FactorAnalysis 因子分析法)。
Text Mining(文本挖掘):
VSM(Vector Space Model向量空间模型),Word2Vec(词向量学习模型),TF(Term Frequency词频),TF-IDF(Term Frequency-Inverse DocumentFrequency 词频-逆向文档频率),MI(MutualInformation 互信息),ECE(Expected Cross Entropy 期望交叉熵),QEMI(二次信息熵),IG(InformationGain 信息增益),IGR(Information Gain Ratio 信息增益率),Gini(基尼系数),x2 Statistic(x2统计量),TEW(TextEvidence Weight文本证据权),OR(Odds Ratio 优势率),N-Gram Model,LSA(Latent Semantic Analysis 潜在语义分析),PLSA(ProbabilisticLatent Semantic Analysis 基于概率的潜在语义分析),LDA(Latent DirichletAllocation 潜在狄利克雷模型)
Association Mining(关联挖掘):
Apriori,FP-growth(Frequency Pattern Tree Growth 频繁模式树生长算法),AprioriAll,Spade。
Recommendation Engine(推荐引擎):
DBR(Demographic-based Recommendation 基于人口统计学的推荐),CBR(Context-basedRecommendation 基于内容的推荐),CF(Collaborative Filtering协同过滤),UCF(User-basedCollaborative Filtering Recommendation 基于用户的协同过滤推荐),ICF(Item-basedCollaborative Filtering Recommendation 基于项目的协同过滤推荐)。
Similarity Measure&Distance Measure(相似性与距离度量):
Euclidean Distance(欧式距离),ManhattanDistance(曼哈顿距离),Chebyshev Distance(切比雪夫距离),MinkowskiDistance(闵可夫斯基距离),Standardized Euclidean Distance(标准化欧氏距离),MahalanobisDistance(马氏距离),Cos(Cosine 余弦),HammingDistance/Edit Distance(汉明距离/编辑距离),JaccardDistance(杰卡德距离),Correlation Coefficient Distance(相关系数距离),InformationEntropy(信息熵),KL(Kullback-Leibler Divergence KL散度/Relative Entropy 相对熵)。
Optimization(最优化):
Non-constrainedOptimization(无约束优化):Cyclic VariableMethods(变量轮换法),Pattern Search Methods(模式搜索法),VariableSimplex Methods(可变单纯形法),Gradient Descent Methods(梯度下降法),Newton Methods(牛顿法),Quasi-NewtonMethods(拟牛顿法),Conjugate Gradient Methods(共轭梯度法)。
ConstrainedOptimization(有约束优化):Approximation Programming Methods(近似规划法),FeasibleDirection Methods(可行方向法),Penalty Function Methods(罚函数法),Multiplier Methods(乘子法)。
Heuristic Algorithm(启发式算法),SA(SimulatedAnnealing,模拟退火算法),GA(genetic algorithm遗传算法)
Feature Selection(特征选择算法):
Mutual Information(互信息),DocumentFrequence(文档频率),Information Gain(信息增益),Chi-squared Test(卡方检验),Gini(基尼系数)。
Outlier Detection(异常点检测算法):
Statistic-based(基于统计),Distance-based(基于距离),Density-based(基于密度),Clustering-based(基于聚类)。
Learning to Rank(基于学习的排序):
Pointwise:McRank;
Pairwise:RankingSVM,RankNet,Frank,RankBoost;
Listwise:AdaRank,SoftRank,LamdaMART;
Tool(工具):
MPI,Hadoop生态圈,Spark,BSP,Weka,Mahout,Scikit-learn,PyBrain…