Hi, everybody. Let’s talk about something interesting today.
When I read Jin Yong’s Swordsman, I always felt that there was something a little ridiculous, namely the swordsmanship of the Huashan school. It is said that some people like to play cheap (sword), some people like to use gas, this is actually a personal preference, how can rise to the door dispute, but also so cannibalism? So I have always felt that this is jin Yong’s plot needs, otherwise how can there be a wind qingyang legend linghu Chong plot?
But as I grew up and entered the workforce, I had a different perspective on this question. Because I also saw a bit of the shadow of the family on the job, although not as bitter as in the novel, but still very interesting, so today and we talk about this topic.
In my observation, algorithmic engineers in the workplace fall into two camps. One is academic school, one is practical school, and the following is a chat with you.
academic
The word academic should be easy to understand, can also be understood as academic. In short, I like to look for solutions or inspiration from academic fields, which can be directly reflected by reading papers.
I find that this faction has something to do with academic degree. The higher the degree, the more formal the academic style of miao Hong. Paper is always referred to, and repetition is always mentioned. The solution often given is that we can follow the methods in a certain paper to try and see if we can solve the current problem. In my daily work, I often take time to read various academic journals and papers.
From the use of a certain model in a certain scene to the processing of a certain feature and sampling, it is inevitable to find the reference and source. In short, all our practices are not groundless, there are traces to follow, can find theoretical evidence. It is a bit like the jianzong in the novel, who values the routine very much and thinks that if the routine is played well, the problem can be solved. On the other hand, losing must be the sword didn’t practice home, or learn the swordsmanship is not bad, it is not the internal failure.
The advantage of this faction is that it looks relatively bright, both in terms of academic background and practice, it looks very high and noble. The language is mixed with English and Chinese, forcing very high. Persuasion is very strong. No matter communicating with laymen or reporting to the boss, as long as you throw out a few terms and the name of a paper, you can intimidate people. Outsiders talk with it, if there is no source cited it is difficult not guilty, feel that their play is wild fox Zen as if not on the hall of elegance.
But the shortcomings are also obvious, will only play tricks actual combat is often very worrying. The reason is simple, because we can find only a few real materials in papers and periodicals. Although the paper will explain the design method of the model and even attach the code, but only these are useless. In the number of storehouse, feature design and processing, these real kung fu things are not revealed. Copying models only works very, very little, and in some extreme cases is harmful.
More importantly, all paper is not systematic. I have written several paper profiles in the past, although they are well-known papers in the recommendation field. However, they are basically limited to the model, and there is no complete introduction and content about the whole recommendation system from top to bottom and from shallow to deep. Personally, I feel that reading a paper gives me a sense of getting a glimpse of a leopard. For Daniel, he already has the whole picture of the leopard in his mind. It is enough to glance at a few key points. But for most less experienced practitioners, it’s almost impossible to see the full picture through this hole.
Practice to send
The opposite of the academic school is the practical school, which can also be simply understood as the wild way.
When I was in Ali, it was obvious that ali’s corporate culture advocated wild ways. You can listen to its slogan, “Ordinary people do extraordinary things”, and “We see because we believe”, so we can get a glimpse of it.
The practice style of the practice is true knowledge, regardless of those rules and regulations. Let me tell you a story of my own to give you a taste of it. When I first came to Ali for a short time, I was just beginning at that time, about the level of reading a few machine learning books. I was asked to predict a user preference category. I did not think about going over to see how the paper was done, or what previous projects had been done, and I did it by myself. I simply designed the scheme and features, most of the features are ready-made, and some of them have some problems in distribution. I made some one-hot or multi-hot processing, and then randomly set up a good model (XGboost).
I felt that I was too hasty, or MAYBE I thought too lofty to make a model at that time. I felt so low that I could not say it out loud, and felt quite ashamed. I still remember when I finished training the first version of the AUC was 0.82. I didn’t think anything was wrong. In fact, the forecast AUC of 0.8 in this scenario is simply off the mark. I had a quick look at it, and it was okay. Take new data to calculate coverage, also very good, most users click on the category hit.
What’s more, AFTER I finished this model, I forgot about it. But to my surprise, from then on, various leaders have been asking me for advice on how to do this model. After comparing their own data, they all think it is very accurate and want to learn more. What is more exaggerated is that later it was said that we were going to make an advertising prediction model, which was specially made by the students of Ali Mom’s advertising algorithm. After a long time, the effect of this model was not as good as it was. When I heard about it, my jaw dropped.
Now looking back at the beginning, although still a little incredible, but calm down to think, there are some reasons. At that time, several values and practices were all set correctly, such as the ratio of positive and negative samples to 1:3, the selection of positive and negative samples, and discretization of some features with uneven distribution, etc., which were all correct. Just at that time basically by feeling and speculation, not as confident as now.
Generally speaking, practitioners like to think of their own methods and design their own solutions to solve problems, rather than blindly referring to paper. For example, if the model effect is not good, the first idea is not to change a model or refer to the paper, but to think about the useful features in this scene, or whether the parameters of the model need to be adjusted. It is a bit like the airbender in Huashan school, controlling the sword with the air, and the theory is determined by practice. The effect is as effective as those tricks and tricks.
I’m thinking
I used to be hands-on, because technology is for the business, in companies large and small. That is to say, there are practical results and outcomes, much more useful than fancy advanced technology. Of course, part of the reason is that I am an undergraduate and have not been baptized and trained in the orthodox academic atmosphere.
Later, as I read more and more papers and had a broader vision, I came up with new ideas on this issue. There are many advanced and clever methods in paper, and if they are abandoned, it is not enough to just do things behind closed doors. In fact, this is not an either-or problem. A good algorithm engineer should not only solve the problem, but also have a good understanding of the development situation and prospects of the industry today. It can not only solve practical problems, but also look forward to the development of the industry, which can be called a major. So now, I think the problem should be divided into thirty and seventy, three points are academic, seven points are doer.
Writing this article is not to point out who is high and who is low, but to provide you with a new vision, look at yourself and compare with others, if you can broaden your vision, find a little resonance, break through the shackles of a little better.
That’s all for today’s article. I sincerely wish you all a fruitful day. If you still like today’s content, please join us in a three-way support.
Click here for more articles