In addition to Weibo, there is also WeChat
Please pay attention
WeChat public account
Shulou
2025-01-29 Update From: SLTechnology News&Howtos shulou NAV: SLTechnology News&Howtos > Network Security >
Share
Shulou(Shulou.com)06/01 Report--
Due to the relationship between work, began to come into contact with the technical field of machine learning, although their work seems to have little to do with machine learning, but the use of machine learning for big data's analysis is very important. Therefore, starting from today, I will start another series of notes on "machine learning", which will mainly record and sort out the gains of my study. Today is a basic introduction to machine learning.
What is machine learning? Why do you need machine learning?
The so-called machine learning, English is Machine Learning, the earliest form is similar to data mining, pattern recognition, artificial intelligence and so on. No matter how it changes and which specific field it is applied to, its core idea has not changed: the use of data analysis technology to discover the rules or patterns hidden behind the data. The main problems here are those whose algorithms are not clear and cannot be solved directly by programming. For example, for the problem of sequence sorting, we already have mature algorithms such as bubbling, but for recognizing handwritten fonts or extracting a person's words from audio, the processing mechanism of the brain is still not clear, so how can we talk about programming? Therefore, for such problems that are difficult to program, we adopt the method of data analysis; and the reason why we can obtain patterns or approximate descriptions through data analysis is essentially because the data itself is not random. It already contains laws that we have not yet discovered, and the purpose of using data analysis is to discover and find these rules or approximate descriptions of these laws. The data analysis here is what we call machine learning today, which allows the computer to learn the algorithm to solve the problem through a large amount of data analysis, so the machine learning algorithm can also be called "learning algorithm".
II. Supervised learning
Next, let's look at four specific aspects of machine learning: supervised learning, algorithm theory, unsupervised learning and reinforcement learning. First of all, let's look at supervised learning. There may be many theories about the definition of supervised learning, but we only need to grasp one thing: supervised learning needs a set of "supervised data" for training. The "supervisory data" here refers to a set of standard output data that are clear about the standard input, that is, the data with clear "correct" results. What we do is to run the algorithm so that the algorithm learns the mapping relationship according to these standard data, so it is "supervised".
The common supervised learning includes regression and classification. The so-called return to the common example is like the data of house prices in different months. The house prices (average price of square meters) vary in 12 months of the year. If we take the time as the x-axis and the average house price as the y-axis, then we can draw a chart of house prices in a year / month. If we want to know the house prices in January next year, we just need to find a curve that best matches the known data, and then use it to predict the average house prices in the future.
As for classification, the most commonly used example is the data on the relationship between benign / malignant and size of the tumor, taking the size of the tumor as the x-axis and the benign / malignant as the y-axis, based on the existing data on the size and nature of the tumor, to judge the nature of the new case. The y-axis here is no longer a continuously changing value (such as the average house price) as in the regression example, but a discrete value (1: benign; 0: malignant). We can also train machine learning algorithms, and then judge new cases according to the rules found. To generalize the idea of classification, now we only consider the size of the tumor, if we add the age of the patient, we will get a three-dimensional function map; if we consider the sex of the patient, then the dimension will continue to increase. If there are many factors to consider in the problems we face, then we may need a high-dimensional plane. But what does this high-dimensional plane look like? At this time, how to find the laws in space? Fortunately, in recent years, people have found SVM to solve the classification problem of this high-order vector space. SVM is now more and more widely used, which will be covered in later chapters.
Third, algorithm theory
Machine learning depends on a variety of algorithms, most of which are based on mathematics and statistics, so understanding these algorithms sometimes requires mastering a certain mathematical foundation. In statistics, such as the expectation, variance and correlation coefficient of random variables, while mathematics requires more algebraic knowledge, such as matrix calculation, transpose, inversion and eigenvector. Generally speaking, if you only use machines to learn existing algorithms, you don't need to have a thorough understanding of the mathematical principles, you just need to understand each algorithm and the conditions under which it is used. Our purpose of learning machine learning is divided into three levels:
-1. Understand the field of machine learning and know important algorithms
-2. We can flexibly apply machine learning algorithms to solve problems according to our own practical problems.
-3. Understand the algorithm and propose the improvement of the existing algorithm
Here my own goal is to the second level, to be able to use machine learning algorithms to solve the data analysis problems. Machine learning is a very useful data analysis tool.
IV. Unsupervised learning
What corresponds to supervised learning is unsupervised learning. The characteristic of unsupervised learning is that at the beginning, there is not a set of standard data that know the results, and the rules are found completely from a pile of clueless data. A common example of unsupervised learning is clustering. A classic example is the "cocktail party" question, which is how you extract the voice of someone you want to hear from the superposition of sounds at a noisy cocktail party. Here, we need a lot of clustering analysis of audio data, which can be used in voice recognition, image pixel analysis, computer vision, social network / market segmentation and so on.
Fifth, enhance learning
The input of the data analysis mentioned above is a large amount of data at once, and then make a prediction / judgment on a new input. But there are some problems that need to be analyzed for an input sequence, that is to say, at this time, what we focus on is not the result of an input, but the "strategy" of an input sequence. For example, in the navigation program of drones, a "rise" or "descent" command will not cause an accident, on the contrary, only a series of continuous take-off and landing instructions will cause a plane crash. The application fields here often focus on the areas of "strategy", such as games (RTS, etc.), UAV, robot navigation and so on.
VI. Summary
It can be said that the field of machine learning is mainly in the three aspects mentioned above: supervised learning, unsupervised learning and reinforcement learning, but the specific application fields are related to medicine, biology, electronic engineering, artificial intelligence and other fields. As a powerful tool, machine learning plays an irreplaceable role in data analysis in different fields.
Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.
Views: 0
*The comments in the above article only represent the author's personal views and do not represent the views and positions of this website. If you have more insights, please feel free to contribute and share.
Continue with the installation of the previous hadoop.First, install zookooper1. Decompress zookoope
"Every 5-10 years, there's a rare product, a really special, very unusual product that's the most un
© 2024 shulou.com SLNews company. All rights reserved.