The company's fourth paradigm
2019.08.17 10:30 side
The interviewer said one side one up interview time is tight, only two hours, so eliminating the need to introduce myself.
Do first a programming problem, in the array are arranged in front of all the even-odd, is not considered in the original order of the array.
This question is not difficult to set up a head and tail pointer, then the cycle, the even hit the head pointer, a tail pointer encountered odd, exchange, know the head pointer is greater than the tail pointer is over.
First, he said his thoughts and then write the code, although there is a small moment suddenly stuck like that, but it is quite well written out.
Then, based on the title, the array change to the list.
In fact, this problem is not difficult, I began to ask a supplementary space can not be used, the interviewer said the best O (1), in fact, is not difficult, then I wanted a.
To traverse the list once, and then know how many even-odd number, then set the two pointers point to the end result of odd and even chain link head head, and traverse, if the odd-even-odd and even chain chain exchange.
However, this operation is quite complex. So the interviewer suggested that I think about time, I thought of another way, in fact, very simple. First traversal, get get a pointer to the last node of the list. Start again from scratch traversing encountered even-numbered nodes, is disconnected to the end of the list, and then on the broken connections. Of course, eventually written out relatively smoothly.
Just right side of the end of two and a half hours of questions.
11:10 two faces
Just after the side that faces today is over, the result is received by the side, ready for a second interview, so after a while entered the two faces.
Two faces do first a simple self-introduction. The following were then asked some questions.
1, maximum likelihood application, gives an example: a city where there are n cars, numbered 1 to n, first arrived in the city, randomly stopped a car, know the numbers x, how do you know n =? Is the total number of vehicles in the city. Using maximum likelihood to solve.
I can think of is the beginning if you know x, for example x = 20, then you can explain at least n> = 20, then p (n <20 | x) = 0. But still want to use is not the maximum likelihood how demand. In fact, maximum likelihood is used to estimate the parameters, such as the known p (x | θ), and then ask the model parameter θ. That is to make p (x | θ) is maximized, because we are here only an experiment, otherwise it is p (x1 | θ) * p (x2 | θ) * ...., maximize it. So here it is actually p (x | n), p (x | n) = 1 / n, because the probability sit every car is the same, but there is only one sample, so to maximize the 1 / n, in fact, n is the smallest, then consider the boundary n> = x, then in fact, the answer came out, stopped to number x, it is estimated that the city's total number of vehicles n = x.
It is clear that the use of maximum likelihood was that the results do not seem good, the interviewer asked why is this happening? No contact with hard to think of the fact, but he slowly guided me think. Bayesian company first asked me, I said posterior probability = prior probability *? , I started to say a formula, he said to say terms, but I forgot to p (x | θ) term, what is it, then he said the likelihood, I recall, the likelihood probability! ! ! Probability Here, naturally leads to a maximum likelihood hypothesis as homogeneous prior probability distribution, i.e., the foregoing n = 1,2,3,4, ... is the same, no difference. But in our practice, a city of the total number of vehicles is regular, that is a certain distribution, such as the number of cars in about obedience class of Gaussian distribution. But using maximum likelihood does not take this into account, so the results obtained above was less accurate.
And then he asked what I mainly projects, mainly interactive game Netease clusters front and defect detection, specifically asked what a little forgotten. Finally, he asked me what the problem is, when asked the extent of their main businesses. The whole process is about 40-50 minutes.
2019.09.01 14:45 on three sides
After two faces separated by two weeks nothing news, that cold, suddenly received a phone about three sides hr while sitting at home.
On three sides by the interviewer smiled and asked me, see my previous internships are in Shenzhen, Guangzhou, Shanghai Why they want to vote, but they are not the kind of big companies, not very well known. Then I said, I can accept Shanghai, Beijing does not want to go, the interviewer smiled and said Shanghai is still south of it, ha ha ha ha. Then chatted project, but did not ask the technical details, but mainly asked business, such as Netease game of clustering project, which is mainly used in the game ah, then the amount of data, then the downstream business and the like, then how to evaluate performance, how to ensure that irregularities are found the studio without mistakenly arrested yet. Netease asked me what games are currently the main game. Then ask how long do kaggle that, then think of how to do this. In short basically did not ask to technology. Finally, asked me any questions, I asked the main business, he introduced what their "prophet" AutoML system, and now there is still AutoCV visual platform. Both platforms seems to have done pretty good, and one day later Huawei hr hr talk time is also used as an example in terms of, he says the industry is doing well. It feels a bit like a pressure surface. Feeling cold, right, Orz, but these have to ask one hundred twenty-three face came closer and technical content. Thank fourth paradigm!