This series of notes is made for learning "Machine Learning" by Mr. Zhou Zhigang. If there are any mistakes, please feel free to advise.
In computer systems, "experience" usually exists in the form of "data". Therefore, the main content of machine learning research is about algorithms that generate "models" from data on computers, that is, "learning algorithms" ( learning algorithm). We provide it with empirical data, and he can generate models based on these data; when faced with new situations, the model will provide us with corresponding judgments.
If computer science is the study of "algorithms", then machine learning is the study of "learning algorithms".
1. Terminology
A collection of records is called a "data set"
Each of these records is a description of an event or object, called an "instance" or "sample"
A matter that reflects the appearance or nature of an event or object in some way, called an "attribute" or "feature"
The value on the attribute, called "attribute value" (attribute value)
The space formed by attributes is called "attribute space" (attribute space), "sample space" (sample space) or "input space"
In this space, each point corresponds to a coordinate vector, so an example is called a "feature vector" (feature vector)
Let D = {x1, x2, x3, .... , xm} denote a data set containing m examples. Each example has d attributes, d becomes the "dimensionality" of the sample xi
The process of learning a model from data is called "learning" or "training", and the data used in this process is called "training data", where each sample is called a "training" The set of training samples is called "training set"
The learned model corresponds to a certain underlying law about the data, called a "hypothesis"; the underlying law itself is called the "truth" or "ground-truth". The learning process is all about finding out or approaching the truth.
In this book, the model is sometimes referred to as a "learner "
Information about the result of an example is called a "label"; an example with label information is called an "example". The ith example is generally represented by (xi, yi) , where yi ∈ Y is the label of xi , and Y is the set of all labels, that is, the “label space” or “output space”
If the prediction is a discrete value (such as "true" or "wrong"), such a learning task is called "classification"; if the prediction is a continuous value (such as a degree of completion 0.95, 0.88), such a learning task is called For "regression" (regression). For "binary classification" tasks involving only two classes, one is usually called the "positive class" and the other is called the "negative class" or "negative class"; involving When there are multiple categories, it is called a "multi-class classification" task.
After learning the model, the process of using it to make predictions is called "testing", and the tested samples are called "test samples".
You can also do "clustering", that is, the samples in the training set are divided into several groups, each group is called a "cluster";
According to whether the training data has label information, learning tasks can be roughly divided into two categories: "supervised learning" and "unsupervised learning". Classification and regression belong to supervised learning, and clustering belongs to unsupervised learning. .
The ability of the learned model to apply to new samples is called "generalization".
It is usually assumed that all samples in the sample space obey an unknown "distribution" (distribution) D, and each sample we obtain is independently sampled from this distribution, that is, "independent and identically distributed" (independent and identically distributed, referred to as iid for short) ). In general, the more training samples, the more information we get about D, the more likely it is to learn a model with strong generalization ability.