That's how I use the data: the warranty Story

At the end of the year, to see 2014 feel or opened a good head, is expected to be the year-end award will not be less, the company is integral, like mobile recharge to send points, in a year I do not have less points! Haha, too witty! Also, I am expected to lead the team to open the second battlefield, a company, processing data a lot, the original is manual operation, I covet them (because my family may look at my blog, I dare not say is "they") for a long time, aimed at a chance to make a day of work, in 16ms to get the results, um , you did not read the wrong "16ms"! Then, "they" asked me not to recruit people, so I won!

There is also a happy thing, I was invited by the Phpchina community as "invited author", say oneself write a blog has one months, period still have a lot of friends contact me, honestly, contact manuscripts out of the book more, but I prefer online community, write a book seems a bit too formal, not ready, Time is also allocated not to come over, you manuscripts friends! Of course, the blog is also an online community, Phpchina is also, I will continue every weekday, then Phpchina will be reproduced in part of the article, in its website, expected follow-up if any activities, I will be the first time to share with you.

Ah! Today's beginning a bit long ah, to sum up, is expected to be very important! Today we are going to talk about the "warranty story" (You may want to say what this has to do with the above! Haha, who cares?)

Now you buy any product has the concept of warranty, because their numbers are too typical and monotonous, so that people think it is to shoot the head out! Common also on "one year", "Three years", "3 months" and so on.

Mapping to software development, which is like some configuration data, you can casually configure, it does not matter, but if you want to do better, it will take a little thought, it is not no way, let us continue!

I asked a question: "For the manufacturers, the warranty costs not"? This question is too simple, must be to spend money! According to my thinking, all the things that want money is a conclusion: "Not a loss"! For example, sell 10, fix 9, that must shut down! If you fix 6, you can be basically flat. If you only fix 3, there is a margin! Excuse me, you are the boss, how many do you want to fix?

Well, the goal is to fix only 3! Follow the train of thought, that is to say: To set a warranty, during this period, only bad 3 or less, the target even achieved! There's still a bit of data here, and there's no way to keep it up, and suppose there's a batch of trial records, marking how long it takes for each product to be bad. Can you imagine, if the general product can be used for 8 months, you say that the day will be a lot of bad? This question is very simple, yes, it should not! So is there a lot of bad in 8 years? It's simple, and it's not that strong! A large number of use records will be gathered in the 8 months near the stall, the distribution is very uneven, two small, the middle of the big, which gives us an estimate caused trouble!

Haha, the evils of mathematics and mathematicians have found answers to such questions, they said: Product damage by a number of factors staggered influence, but their working time in line with the "Normal distribution" (self-Baidu), as long as the average and standard deviation, and then set a formula, then the answer has! I'm too lazy to remember the formula, the result is about this (U is average, x is standard deviation)

1, less than 0.03% of u+3x or less than u-3x

2, there are 95.4% between u-2x to U+2X

3, there are 68.3% between U-x to U+x

I know you can not remember this thing, because I also can't remember, but as a traitor to the mathematics department, I have long thought of the loss of strokes, especially for programmers, let's look at

1, call your familiar programming language random number generator, generate a lot of random numbers between 0 to 1, more points for good, such as millions of

2, can be generated on one side of the standard deviation, the average value is not to forget, is 0.5

3, then count, want to know the percentage of u-1.2x to u? In this random number, the number of sub miles can be completely in line with the real world of any other event, do not believe? You can try!

This is almost a universal tool, any confused, the more confused the idea of the thing, the more effective! It's easy to say, but I said data-driven doesn't need mathematical theory, although it's simple, let's go!

Suppose your product starts selling, sells a lot, also collects a lot of product maintenance record, then can adopt the Invincible method that data drives advocate: "What happens is the fact, what happens is the truth"! Here, for example, to estimate the percentage of damage in 1 years, the damage in the historical data within a few years, and then divided by the total sold in a year, the problem solved, as long as you have enough data, and the theoretical method is almost the same, very strange, like doomed!

At this point, no matter how much the boss wants to accept the maintenance percentage, your program can give a back-push warranty. This is also the most knowledgeable enterprise real use of the development of the warranty method Oh!

Well, the story is about to end, according to the usual way of thinking about the data. As mentioned above, we emphasize the effect of a reverse push, this allows our program to foresee a lot of things, before those things happen to give a conclusion, such conclusions can be used in a lot of places, the program's early warning, business alerts, abnormal operation of the prompts and so on, do not think it is indifferent, in fact, you think, If a program can't appreciate any obvious results from the data, does it look like an idiot?

is expected to be important!

Come on, buy my bike, 143 days warranty!

See more highlights of this column: http://www.bianceng.cnhttp://www.bianceng.cn/Programming/project/