An introduction to Statistical Learning - JXXCgit/Notes-Learning-for-Data-Scientists GitHub Wiki

Reference web: https://www.statlearning.com/

Chapter 13 Multiple Testing

Chapter 10 Deep Learning

Chapter 12 Unsupervised Learning

Chapter 9 Support Vector Machines

Chapter 8 Tree-Based Methods

Chapter 6 Linear Model Selection and Regularization

Chapter 4 Classification

Chapter 5 Resampling Methods

Chapter 7 Moving Beyond Linearity

Chapter 11 Survival Analysis and Censored Data

Chapter 3 Linear Regression

Chapter 2 Statistical learning

* how to differentiate 'prediction' and 'inference' problem in the modeling?

**### ### Inference is when you use that model to learn about things that happened in the past. However, prediction is when you use your model to make predictions in the future. **

Inference: i.e., which media are associated with sales?/ which media generate the biggest boost in sales? / how large of an increase in sales is associated with a given increase in TV advertising? or how much extra will a house be worth if it has a view of the river?

Prediction: i.e., is the house under or over-valued? Linear model usually allow for relatively simple and interpretable inference, but not yield as accurate prediction as other approaches. while highly non-linear models can potentially provide quite accurate prediction, but comes at the expense of a less interpretable model

* how to estimate ƒ?

Parametric method: based on assumptions (model-based) and then use the training data to fit/train the model. i.e., using least squares is one of many ways to fit the linear model

Non-Parametric method: not based on explicit assumptions. i.e., thin-plate spline to estimate ƒ

Non-parametric method can accurately fit a winder range of possible distribution for ƒ than parametric way

Trade-Off between Prediction Accuracy and Model Interpretability