My CSDN blog column: https://blog.csdn.net/yty_7
Github address: https://github.com/yot777/
Hello everyone, thank you for your attention to the column of Yang Tao on the CSDN blog.
After passing the basic Python tutorials and advanced lectures, I believe you must have a general understanding of Python, a programming language, and a perceptual understanding through some examples.
My tutorials and lectures are just an outline, and there are many details of Python that are not comprehensive. Please promote your own spirit of self-sufficiency. If you do n’t understand, just Baidu and CSDN!
Ok, let's get back to business, we start a new chapter-machine learning.
What is "learning" and "learning"
Please note that my titles are "Learning" and "Learning", not learning. why? Because in ancient Chinese, "learning" and "habit" are two completely different words. In modern times, "learning" and "habit" have become the word "learning".
So what do "learning" and "learning" mean? A sentence from the sage Confucius' Analects is the best answer.
Confucius said, "Do you not only learn when you learn?" (If you haven't learned this sentence for half an hour)
Translation: Confucius said: "Isn't it a delightful thing to review and practice on time after learning the knowledge?"
The sage is the sage, and the "learning" and "learning" things are so clear and concise: three key points: learning knowledge , reviewing on time , and practical exercises .
Recall that when you were in school, did you come here that way?
Learned knowledge: The teacher talks about the solution method of the unary quadratic equation in the classroom, and derives the formula:
Review on time: follow the book to understand the meaning of the formula after class
The formula is memorized in my mind through many revisions. If the memory is not correct, repeatedly recall the derivation process.
Practical exercises: Apply the formulas memorized in your mind to practical calculations during the exam. If the memory is not correct at this time, it will affect the test scores, and the price will be a bit high.
We are very clear about our "learning" and "learning" process, what about machines? Exactly similar!
Learned knowledge: Mathematical modeling through a large amount of training data allows the machine to "learn" a certain pattern of data.
Review on time: run the test data through the model that has been built (the amount of test data is generally 20% to 30% of the training data) to verify whether the laws learned are correct. If correct, enter the practice environment. If it is not correct, return to the first step and let the machine continue to "learn". Note: Correctness generally does not refer to 100% complete correct rate, but there is a threshold, which varies according to the application scenario, such as 99%, 95%, 90%, etc. Once this threshold is reached, it indicates that the achieved effect can be applied to practice.
Practical exercise: Apply the model that the machine "learns" to the data to the actual scene to solve the problem, such as classification, prediction, clustering, etc. If there is a situation where the correct rate threshold is not reached in the practice link, it should be regarded as a program defect, and the first step needs to be retrained and modeled.
Why use machines to learn
Before explaining this question, let's compare the significant differences between human learning and machine learning:
(1) Due to the diversity and complexity of the original data, humans are far less capable of recognizing new things than machines, and many mathematical modeling manuals are simply impossible.
(2) Because the subjective state of human beings fluctuates greatly, under the influence of physical, mental, emotional and other factors, people cannot remember the verified models under any circumstances, such as the root-finding formula of the quadratic equation
It is quite possible to forget about an on-site exam in a bad state. The machine is different. In any case, as long as you need it, it can accurately use the verified model.
(3) The machine does not have subjective colors, and it is all programmed input and output, which can be truly "unselfish." Humans have never been able to do this, just like the referees in many sports games score, there is always a part of "impression score", which actually has room for action. You know.
Machine learning is not a panacea. Remember, machines can replace some human thoughts, but they can never replace human emotions. Reason + emotion is the complete human. But we must have a crisis in advance, beware of a human tragedy like "Terminator", we must put a strong shackle on the machine so that it can only serve humanity under human control.
to sum up
"Study" is the meaning of Kong Shengren's "learning while studying". There are three key points: learning knowledge , reviewing on time , and practical exercises .
The learning steps of machines and humans are completely similar, with the same three keywords: training data , test data , and actual scenarios .
Machine learning has significant advantages over humans, but the machine must be firmly yoke so that it can only serve humans under human control.
My CSDN blog column: https://blog.csdn.net/yty_7
Github address: https://github.com/yot777/
If you think this chapter is helpful to you, welcome to follow, comment and like! Github welcomes your Follow and Star!
