首页 > 代码库 > scikit-learn:在实际项目中用到过的知识点(总结)

scikit-learn:在实际项目中用到过的知识点(总结)

零、全部项目通用的:

http://blog.csdn.net/mmc2015/article/details/46851245(数据集格式和预測器)


http://blog.csdn.net/mmc2015/article/details/46852755(载入自己的原始数据)

适合文本分类问题的 整个语料库载入)


http://blog.csdn.net/mmc2015/article/details/46906409(5. 载入内置公用的数据)

(常见的非常多公共数据集的载入,5. Dataset loading utilities)


http://blog.csdn.net/mmc2015/article/details/46705983(Choosing the right estimator(你的问题适合什么estimator来建模呢))

一张图告诉你,你的问题选什么estimator好。再也不用试了)


http://blog.csdn.net/mmc2015/article/details/46857949(训练分类器、预測新数据、评价分类器)


http://blog.csdn.net/mmc2015/article/details/46858009(使用“Pipeline”统一vectorizer => transformer => classifier、网格搜索调參)





一、文本分类用到的:

http://blog.csdn.net/mmc2015/article/details/46857887(从文本文件里提取特征(tf、idf))

CountVectorizerTfidfTransformer


http://blog.csdn.net/mmc2015/article/details/46866537(CountVectorizer提取tf都做了什么)

深入解读CountVectorizer都做了哪些处理。指导我们做个性化预处理


http://blog.csdn.net/mmc2015/article/details/46867773(2.5.2. 通过TruncatedSVD实现LSA(隐含语义分析))

LSALDA分析


(非scikit-learn)http://blog.csdn.net/mmc2015/article/details/46940373(《textanalytics》课程简单总结(1):两种word relations——Paradigmatic vs. Syntagmatic)

(非scikit-learn)http://blog.csdn.net/mmc2015/article/details/46941367(《textanalytics》课程简单总结(1):两种word relations——Paradigmatic vs. Syntagmatic(续))

(词粒度关系:Paradigmatic(聚合关系:同性质可相互替代、用基于tfidf的相似度挖掘) vs. Syntagmatic(组合关系:协同出现、用互信息挖掘))


(非scikit-learn)http://blog.csdn.net/mmc2015/article/details/46771791(特征选择方法(TF-IDF、CHI和IG))

(介绍了TF-IDF在特征选择时的误区、CHI Square和Information Gain在特征选择时的应用





二、数据预处理用到的(4. Dataset transformations)

http://blog.csdn.net/mmc2015/article/details/46991465(4.1. Pipeline and FeatureUnion: combining estimators(特征与预測器结合;特征与特征结合))

特征与预測器结合、特征与特征结合


http://blog.csdn.net/mmc2015/article/details/46992105(4.2. Feature extraction(特征提取,不是特征选择))

loading features form dicts、feature hashing、text feature extraction、image feature extraction


http://blog.csdn.net/mmc2015/article/details/46997379(4.2.3. Text feature extraction)

text feature extraction


http://blog.csdn.net/mmc2015/article/details/47016313(4.3. Preprocessing data(standardi/normali/binari..zation、encoding、missing value))

Standardization, or mean removal and variance scaling(标准化:去均值、除方差)、Normalization(正规化)、Feature Binarization(二值化)、Encoding categorical features(编码类别特征)、imputation of missing values(归责缺失值))


http://blog.csdn.net/mmc2015/article/details/47066239(4.4. Unsupervised dimensionality reduction(降维))

PCA、Random projections、Feature agglomeration(特征集聚))


http://blog.csdn.net/mmc2015/article/details/47069869(4.8. Transforming the prediction target (y))

Label binarizationLable encoding(transform non-numerical labels to numerical labels)



三、其它重要知识点:

http://blog.csdn.net/mmc2015/article/details/47099275(3.1. Cross-validation: evaluating estimator performance)

交叉验证


http://blog.csdn.net/mmc2015/article/details/47100091(3.2. Grid Search: Searching for estimator parameters)

搜索最佳參数组合


http://blog.csdn.net/mmc2015/article/details/47121611(3.3. Model evaluation: quantifying the quality of predictions)
模型效果评估:score函数、confusion matrix、classification report等


http://blog.csdn.net/mmc2015/article/details/47143539(3.4. Model persistence)

保存训练好的模型到本地joblib.dump & joblib.load pickle .dump & pickle .load




None、经常使用的监督非监督模型

http://blog.csdn.net/mmc2015/article/details/46867597(2.5.  矩阵因子分解问题)


http://blog.csdn.net/mmc2015/article/details/47271039(scikit-learn(project中用的相对较多的模型介绍):1.4. Support Vector Machines)

SVM(SVC、SVR


http://blog.csdn.net/mmc2015/article/details/47271195(scikit-learn(project中用的相对较多的模型介绍):1.11. Ensemble methods)

Bagging meta-estimator、Forests of ranomized trees、AdaBoost、Gradient Tree Boosting(Gradient Boosted Regression Trees (GBRT) )


http://blog.csdn.net/mmc2015/article/details/47333499(scikit-learn(project中用的相对较多的模型介绍):1.12. Multiclass and multilabel algorithms)

Multiclass classification、Multilabel classification、Multioutput-multiclass classification and multi-task classification


http://blog.csdn.net/mmc2015/article/details/47333579(scikit-learn(project中用的相对较多的模型介绍):1.13. Feature selection)

Univariate feature selection(单变量特征选择)、recursive feature elimination(递归特征消除)、L1-based / ree-based features selection(这个也用的比价多)、Feature selection as part of a pipeline


http://blog.csdn.net/mmc2015/article/details/47333839( 

scikit-learn(project中用的相对较多的模型介绍):1.14. Semi-Supervised



http://blog.csdn.net/mmc2015/article/details/47414271(scikit-learn(project中用的相对较多的模型介绍):2.3. Clustering(可用于特征的无监督降维))







scikit-learn:在实际项目中用到过的知识点(总结)