3.2 Basic machine learning algorithms

Machine learning can be divided into several types according to different computational results. These different purposes determine that machine learning can be divided into different models and classifications in practical applications.

As mentioned earlier , machine learning is a Cross - disciplinary subject in many fields and a new subject in many fields. , therefore , it will use in practice the classical research methods in different disciplines, namely the algorithm.

3.2.1MachineLearningThe algorithmProcess

The first thing to know is that for machine learning, a machine learning process is a complete project cycle, which includes data acquisition , feature extraction and classification of data, and subsequent algorithms to Create a machine learning model to obtain predictive data. The entire machine learning algorithm is shown in flow 3-2.

Figure 3-2 Algorithm flow for machine Learning

In a complete machine learning process, the entire machine learning program uses the data to create a Learn "model". This model can dynamically adjust and feedback itself, so as to better classify and process the unknown data.

A complete Machine Learning Project consists of the following :

L input data: A naturally acquired data set containing identified and not identified parts as the most basic part of machine learning .

L Feature Extraction: Extract the eigenvalues of the data in a variety of ways. Generally speaking, the more data is included, the more accurate the machine learning model is, and the more difficult it is to handle. therefore , it is very important to find a balanced point of the characteristic size properly.

L Model Design: Model design is the most important part of machine learning, according to the existing conditions, the choice of different classifications, the use of different indicators and techniques . The training of the model relies more on the collection of data and the extraction of features, which requires the support of the above sections.

L Data prediction: Through the knowledge and use of the trained mode, the learning machine can be used to study the methods, theories and techniques of developing, simulating and expanding human's multiple intelligences.

The entire machine learning process is a complete project lifecycle , with each step being based on a previous step .

3.2.2classification of basic algorithms

Depending on the input data and the processing requirements of the data, machine learning chooses different kinds of algorithms to train the model . The choice of algorithm training has no specific pattern, in general, only need to consider the input data form and complexity and user model experience, then the algorithm training, so as to obtain better learning results.

based on the training mode of the basic algorithm , the algorithm can be divided into the following categories (Figure 3-3):

learning : completely black box training of a training method, for input data in the run Span style= "font-family: Italic _gb2312" > There is no difference before the end and identity . Entirely by machine data are identified forming the Span style= "font-family: Italics _gb2312" "Unique analysis model Training Process absolutely no guidance analysis .

data is artificially classification, was artificially Mark and identify. continuously revise and refine the model by learning from the data identified by the human model correctly classify data after a given for classification.

l  Semi-supervised learning: create same model analyze data and identify it algorithm run between supervised and unsupervised finally make all input data can be . Half supervised learning main for eigenvalue missing data analysis .

L Intensive Learning: Learn , feedback , and Modify existing models by entering different identity data, using a machine learning data model that you already have. A new model algorithm is established to identify the input data.

different algorithms have different purposes and requirements. Machine Learning has many algorithms to choose from when it is actually used , and there are many modifications and changes to the different algorithms. for A particular problem, it is difficult to select an algorithm that conforms to the data rule .

Generally used more is supervised learning and unsupervised learning, but due to the popularization of big data, more data will produce a large number of missing eigenvalues, so the next period of time, semi-supervised learning gradually become Hot up .

3.2 Basic machine learning algorithms