A computer program is said to be learnable when it is capable of learningTask Twith someonePerformance indexP,fromExperienceEmiddle school study. The characteristic of this kind of learning is that its performance on T, measured by P, will improve with the increase of experience E.
Task T: Write a computer program to identify chrysanthemums and roses
Experience E: A lot of pictures of chrysanthemums and roses (i.e., training samples)
Performance indicator P: Different machine learning algorithms will be different (for example, the recognition rate can be used as the performance indicator, that is, more chrysanthemums are recognized as chrysanthemums, and more roses are recognized as roses. The correct rate of recognition is called the recognition rate)
According to the definition, machine learning is to construct a certain algorithm for the task of identifying chrysanthemums and roses. The characteristic of this algorithm is that when more and more pictures of chrysanthemums and roses are trained, that is, when there are more and more E , the recognition rate P will also become higher and higher.
1.2 Characteristics of machine learning
data = features
answer=label
Rules = Model
Sample = data + answer
1.3 Basic concepts about data
The data format processed is the same as excel table or structured data table.
The entire table is called a data set (that is, the set of all samples)
Rows are called samples or instances
Columns are called features or attributes
2. Main work of machine learning
According to the learning method: supervised learning, unsupervised learning, reinforcement learning, etc.
According to learning tasks: classification, regression, clustering, dimensionality reduction, etc.
2.1 Learning from data [Supervised learning]
A common supervised learning task isClassification and regression。
When a model is used to predict the category of a sample, it is a classification problem, for example, to distinguish whether the object in a given picture is a cat or a dog.
When the sample results to be predicted are continuous values, it is a regression problem, such as predicting the market price of a stock in the next week.
2.2 Analyze new problems without experience [Unsupervised learning]
Unsupervised learning (some books are also called unsupervised learning) is when there is no training set.For data without labelsA way to conduct analysis and build appropriate models to give solutions to problems.
In unsupervised learning, samples do not have corresponding labels or target values.
Semi-supervised learning is a combination of supervised learning and unsupervised learning.
Commonly used in unsupervised learning tasks areClustering and dimensionality reduction