Learn Zhou Zhihua's machine learning essays
Chapter One Introduction
Model: generally refers to the results learned from data
1.2 Basic terminology
Machine learning requires data, and the set of records is called a data set.
A record is a description of an event or object, also called an example or sample.
As a thing, a sample has its attributes. The value of the attribute is called the attribute value, and the space formed by the attribute is called the attribute space.
To train a model, you need labeled data, which is called a sample.
y is a collection of labels, called label space or output space.
After the model is built, the process of predicting it is called "testing", and the sample being tested is called "testing sample".
The learning task of predicting discrete values, such as predicting good or bad, is called classification. At the same time, "classification" is also divided into two classifications and multi-classification.
The learning task of predicting continuous values is called regression.
The essence of the prediction task is to learn from the training set so that a mapping from the input space X to the output space Y can be established.
Clustering learning: By learning unlabeled data sets, it helps us understand the internal laws of the data. According to a certain standard (such as distance), a data set is divided into different classes or clusters, so that the similarity of the data objects in the same cluster is as large as possible, and there is no difference between the data objects in the same cluster. Sex is as great as possible.
The ability of the learning model to apply to new samples is called generalization ability.
1.3 Hypothetical space
Hypothesis space: The space composed of all hypotheses in the learning process.
Version space: A set of hypotheses consistent with the training set.
1.4 Inductive preference
Occam's razor principle: If there are multiple hypotheses that are consistent with observations, choose the simplest one.