python: Basics of using scikit-learn ①
This time it will be the heart of machine learning I will post about scikit-learn.
Overview of scikit-learn
Read as Cykit Learn. A Python machine learning library.
Cheat sheet
About the analysis (classification, regression, etc.) that you want to perform A cheat sheet that allows you to easily select a model.
Classification
Determine which class it belongs to.
SGD（stochastic gradient descent） For more than 100,000 data Linear classification method
Kernel approximation
For more than 100,000 data If SGD doesn't work Non-linear classification method
Linear SVC For less than 100,000 Linear classification method
k-nearest neighbor method
For less than 100,000 If Linear SVC doesn't work This is a non-linear classification method.
Naive bayes
For text data
Regression
Prediction of target value
SGD（stochastic gradient descent） For more than 100,000 data Linear regression analysis method
LASSO、ElasticNet For less than 100,000 When some of the explanatory variables are important Regression analysis method
Ridge、Liner SVR For less than 100,000 When all the explanatory variables are important Regression analysis method
SVR (Gaussian kernel), Ensemble
If Ridge or Liner SVR doesn't work Non-linear regression analysis method
Clustering
Things to divide according to some rules
KMeans When it is possible to decide in advance how many clusters to divide into Clustering analysis method
MiniBatch For more than 100,000 data A method of learning while dividing data
Spectral clustering, GMM
If KMeans doesn't work Non-linear clustering analysis method.
MeanShift、VBGMM When it is not possible to decide in advance how many clusters to divide into It is a clustering analysis method.
Other
Dimensionality reduction
In the pretreatment process We will do it to improve learning efficiency
PCA, kernel PCA, Isomap, Spectral Embedding, etc.
Hyperparameter optimization
Adjustment values such as learning methods are called "hyperparameters".
There are methods such as grid search and cross validation.
Recommended Posts