p3-Regression-Case Study

Regression:

• Stock Market Forecast
• Self-driving Car
• Recommendation
• Estimating the Combat Power of a pokemon after evolution

Function:
f( pokemon ) = CP after evolution

Method:
Step1:Model
Linear model：f1: y = b + w * xcp (w and b are parameters，xcp is CP before evolution)【w is weight, b is bias】
Another Model：y = b + w1 * xcp + w2 * xcp^2
…
Redesign the model：different species have different function（引入seg函数）
…
Step2:Goodness of Function
Training Data
Loss function L：L(f) = L(w,b)【how bad a funtion is（此处为方差）】
Regularization：
Step3:Best Funtion
A set of function -> Goodness of function <- Training Data
Gradient Descent：偏微分

Overfitting：
A more complex model yields lower error on training data.
A more complex model does not always lead to better performance on testing data.