Statistical Learning Methods Chapter 1 learning experiences

The first chapter of the statistical learning method may be inappropriate. Please stop and correct your learning experience. Thank you first. The previous post said that three elements of statistical learning methods: model, strategy, algorithm, model: Model hypothesis Space Policy: model selection criteria, that is, the method for defining the optimal model. Algorithm: an algorithm for model learning, that is, an algorithm for solving the optimal model. --------------------------------------------------------------------- In general, the application statistics learning method has the following steps: 1. first, we need to analyze existing data and select an appropriate (with unknown parameters) equation or function that can describe the rule of the dataset. This is the first step, it is also an important step. ---> This is also called the hypothetical space of the model. 2. then, because the function (also called the hypothetical space of the model) in step 1 still carries unknown parameters, and its prediction capability is unknown, therefore, we need a method to automatically evaluate the merits and demerits, that is, under what circumstances should we consider that the selection of unknown parameters is good and under what circumstances is bad, in this way, we have a judgment standard, and we can find a way to make the results more and more realistic. We call it a policy (model selection criteria ). 3. finally, we have a model (a function with parameters and a hypothetical space) and a judgment standard. Of course, the dataset is always there, in this way, we can use the amazing, fast, repetitive, and never-tired computing power of the computer to find the optimal parameters of the model (whether or not they are optimal depends on the policy ). In this way, we can predict and classify the data by solving the model. Then, another question arises: how to assess the model prediction capability, and how to make the best use of a limited dataset when the dataset is insufficient? 1. in the Model Solving Process in the previous step, we have a loss function to describe the model policy and judge whether the model is good or bad, in this case, how can we evaluate the acquired model prediction capabilities? Because we are blindly pursuing the ability to predict training data, the model we choose has too many parameters, and the complexity is higher than that of the real model, this has resulted in a special phenomenon: the model has good prediction ability for training data, but has poor prediction ability for unknown data. Therefore, we need a method to assess the model's ability to predict unknown data, rather than simply focusing on the prediction capability when the model is determined. The generalization ability is used to describe the model's ability to predict unknown data. 2. cross-validation of efficient use of datasets