Datawhale-李宏毅机器学习-task02

estimate

The P3 vedio gives a different view to regression model. Compared with the traditional regression model, an important difference is that the regression section of ML course is based on gradient descent, while the traditional regression analysis solve the matrix equation to get the estimators.

problems in real sense

When calculating the estimators of only 2 parameters. We encounter a problem that the best solution is not reached. Here a new method is applied but hasn’t been explained.

model selection

Altert overfitting! We shall find the model that is just suitably complex.

regularization

Add some new terms into the loss func. The added part is used to punished the parameters, which smooth the curve.

The larger $\lambda$, the more we consider the size of w’s and less the error. When regularizing, there is no need to consider the bias.