Shulou(Shulou.com)06/03 Report--

Machine learning algorithms are complex systems that need to be understood through research. Learning the static description of the algorithm is a good start, but it is not enough for us to understand the behavior of the algorithm. We need to understand the algorithm dynamically.

Machine learning algorithm

The running experiment of machine learning algorithm will make you draw experimental conclusions for different types of problems and have an intuitive understanding of the causal relationship between experimental conclusions and algorithm parameters. In this article, you will know how to study and learn a machine learning algorithm. You will learn five simple steps that you can use to design and complete your first machine learning algorithm experiment, and you will find that machine learning experiments are not only exclusive to scholars, but also to you; you will also know that experiments are the only way to proficiency, because you can learn causality from experience, which cannot be learned anywhere else.

What is the research of machine learning algorithm

When studying a machine learning algorithm, your goal is to find the machine algorithm behavior that can get good results, and these results can be extended to multiple problems or types of problems. You do systematic research on the state of the algorithm to study learning machine learning algorithms. This work is accomplished by designing and running controllable experiments. Once you have completed an experiment, you can explain and submit your conclusions. These conclusions will give you a glimpse of causality in algorithmic changes. This is the relationship between the behavior of the algorithm and the conclusions you get.

How to study Learning Machine Learning algorithm

In this section, we will learn five simple steps through which you can study and learn a machine algorithm.

1. Select an algorithm

Choose an algorithm that you have a problem with, or you may find that it performs well in other environments, and you want to use it in the future. For the purpose of the experiment, it is helpful to use an off-the-shelf algorithm. This will give you a bottom line: bug is the least likely to exist. Implementing an algorithm yourself may be a good way to understand the process of the algorithm, but during the experiment, additional variables, such as bug, and a large number of micro decisions that must be made for the algorithm are introduced. Welcome to join big data Learning Exchange and sharing Group: 658558542 blow water exchange and study together (click on ☛ to join the group chat)

two。 Identify a problem

You must have a research question that you are trying to find an answer to. The more specific the question, the more useful it is. The example questions given include the following aspects:

In KNN algorithm, what is the effect of K value as a part of sample space when it increases?

In SVM algorithm, what is the influence of choosing different kernel functions on binary classification problem?

In the binary classification problem, what is the influence of the scaling of different parameters in logical regression?

In the random forest model, what is the effect of adding arbitrary attributes to the training set on the classification accuracy?

For the algorithm, design the questions you want to answer. Think about it carefully, then list 5 evolving questions, and delve deeper into which one is the most accurate.

3. Design experiment

Pick out the key elements from your questions and make up your experiment. For example, take the above example question: "what is the impact of different parameter scaling on logical regression in the binary classification problem?"

The elements you pick out from this question to design the experiment are:

Attribute scaling: you can use methods such as normalization, standardization, raising an attribute to the multiplier, taking logarithms, and so on.

Logical regression: what kind of logical regression do you want to use?

Binary classification problem: there are criteria for binary classification problems with different numerical attributes. A variety of questions need to be prepared, some of which are of the same size (like the ionosphere), while others have different scaling values (such as diabetes).

Performance: model performance scores similar to classification accuracy are required

Take the time to carefully select the elements of your question to give the best answer to your question. Welcome to join big data Learning Exchange and sharing Group: 658558542 blow water exchange and study together (click on ☛ to join the group chat)

4. Conduct experiments and report your conclusions

To complete your experiment, if the algorithm is random, you need to repeat the experiment many times and record an average and standard deviation. If you are looking for differences in results between different experiments (such as with different parameters), you may want to use a statistical tool to indicate whether the differences are statistically significant (like students't-tests), some tools such as R and scikit-learn/SciPy to complete these types of experiments, but you need to put them together and write scripts for the experiments. Other tools such as Weka have a graphical user interface. The tools you use don't affect the problems and the rigor of your experimental design, and summarize your experimental conclusions. You may want to use charts. It is not enough to present the results alone, they are just numbers. You have to associate numbers with problems and extract their meaning through your experimental design.

What does the experimental result imply to the experimental question?

Remain skeptical. What kind of loopholes and limitations are left in your conclusion? Don't run away from this part. Knowing the limitations is as important as knowing the experimental results.

5. Repetition

Repeat operation

Continue to study the algorithm of your choice. You even want to repeat the same experiment with different parameters or different test data sets. You may want to deal with the limitations of your experiment, not just on one algorithm, but to build knowledge and intuition about algorithms by using some simple tools to ask good questions and maintain rigour and skepticism. Your understanding of the behavior of machine algorithms will soon reach a world-class level. Learning algorithms are not only for scholars to do, you can also learn to study machine learning algorithms.

You don't need a high degree, you don't need to train in a research way, and you don't need to be a scholar. For everyone with computers and strong interests, the systematic study of machine learning algorithms is open. In fact, if you major in machine learning, you will certainly adapt to the systematic study of machine learning algorithms. Knowledge will not come out by itself at all. You need to rely on your own experience to get it.

When talking about the applicability of your findings, you need to be skeptical and cautious. You don't necessarily ask unique questions. You will also gain a lot by studying general problems, such as summarizing the general impact of a parameter based on some general standard data sets. You may find the limitations or even counterexamples of some common examples with optimal methods.

Action steps

Through controllable experiments, you know the importance of studying the behavior of learning machine learning algorithms. You have mastered five simple steps and you can design and run your first experiment on a machine learning algorithm to take action. Use the steps you learned in this blog post to complete your first machine learning experiment. Once you have completed one, or even a small one, you will gain the confidence, tools, and ability to complete the second and more.

Conclusion

Thank you for watching. If there are any deficiencies, you are welcome to criticize and correct them.

If you have a partner who is interested in big data or a veteran driver who works in big data, you can join the group:

658558542 (click on ☛ to join the group chat)

It collates a large volume of learning materials, all of which are practical information, including the introduction to big data's technology, high-level analysis language for massive data, distributed storage for massive data storage, and distributed computing for massive data analysis. for every big data partner, this is not only a gathering place for Xiaobai, but also Daniel online solutions! Welcome beginners and advanced partners to join the group to learn and communicate and make progress together!

Finally, I wish all the big data programmers who encounter bottlenecks to break through themselves and wish you all the best in the future work and interview.

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.