A more efficient developer: disable notifications
You don't even want to see the number of unread emails in your taskbar.
In fact, you don't need to see any notifications. Disable them all.
Don't let them break your concentration.
Read full article from A more efficient developer: disable notifications
Temporary Tables: Local vs. Global | Software Development content from SQL Server Pro
A local temporary table, #table_name, exists only for the duration of a user session or the procedure that created the temporary table. When the user logs off or when the procedure that created the table completes, the local temporary table is lost. Multiple users can't share a local temporary table because it is local to one user session. You also can't grant or revoke permissions on the local temporary table. (See also, "Table Variables vs. Temporary Tables").
Read full article from Temporary Tables: Local vs. Global | Software Development content from SQL Server Pro
Notes on Distributed Systems for Young Bloods – Something Similar
New systems engineers will find the Fallacies of Distributed Computing and the CAP theorem as part of their self-education. But these are abstract pieces without the direct, actionable advice the inexperienced engineer needs to start moving1. It's surprising how little context new engineers are given when they start out.
Below is a list of some lessons I've learned as a distributed systems engineer that are worth being told to a new engineer. Some are subtle, and some are surprising, but none are controversial. This list is for the new distributed systems engineer to guide their thinking about the field they are taking on. It's not comprehensive, but it's a good beginning.
The worst characteristic of this list is that it focuses on technical problems with little discussion of social problems an engineer may run into. Since distributed systems require more machines and more capital, their engineers tend to work with more teams and larger organizations. The social stuff is usually the hardest part of any software developer's job, and, perhaps, especially so with distributed systems development.
Read full article from Notes on Distributed Systems for Young Bloods – Something Similar
Please stop calling databases CP or AP — Martin Kleppmann's blog
In his excellent blog post Notes on Distributed Systems for Young Bloods, Jeff Hodges recommends that you use the CAP theorem to critique systems. A lot of people have taken that advice to heart, describing their systems as "CP" (consistent but not available under network partitions), "AP" (available but not consistent under network partitions), or sometimes "CA" (meaning "I still haven't read Coda's post from almost 5 years ago").
I agree with all of Jeff's other points, but with regard to the CAP theorem, I must disagree. The CAP theorem is too simplistic and too widely misunderstood to be of much use for characterizing systems. Therefore I ask that we retire all references to the CAP theorem, stop talking about the CAP theorem, and put the poor thing to rest. Instead, we should use more precise terminology to reason about our trade-offs.
Read full article from Please stop calling databases CP or AP — Martin Kleppmann's blog
Planning the Plan <= :julianbrowne
Wherever you are in the world and whatever business you are in, 2009 is going to be a tougher year than usual. If you aren't at the pointy end of the economic downturn then your customers will be. For a lot of IT departments that's going to mean less to spend. For the lucky ones without budget cuts there's going to be a higher expectation of return on that spend.
Read full article from Planning the Plan <= :julianbrowne
The Death of Architecture <= :julianbrowne
If you looked for places on the earth to experience unusual clarity of thought I don't suppose the island of Aruba would make it to your shortlist. I recently got back from a trip there to watch my brother-in-law get married, but if you're not from the North or South Americas then you probably haven't even heard of it (it being a very long multi-hop flight from everywhere else). I certainly hadn't and had to fly there from London via Miami. Fourteen hours in the air, plus six or so hours hanging around the airport. And let me say this - I hate flying. I hate everything about it. The pre-dawn parking, the waiting, the lugging heavy bags, the faint medical smell of the airport, but most of all the flying. Despite a total acceptance of the statistical fact that I am more likely to die on the way to the airport than in the sky, I have an all-encompassing fear of flying, one which recent membership to the Lorazepam Flight Club has done little to abate.
Read full article from The Death of Architecture <= :julianbrowne
Brewer's CAP Theorem <= :julianbrowne
On Friday 4th June 1976, in a small upstairs room away from the main concert auditorium, the Sex Pistols kicked off their first gig at Manchester's Lesser Free Trade Hall. There's some confusion as to who exactly was there in the audience that night, partly because there was another concert just six weeks later, but mostly because it's considered to be a gig that changed western music culture forever. So iconic and important has that appearance become that David Nolan wrote a book, I Swear I Was There: The Gig That Changed the World, investigating just whose claim to have been present was justified. Because the 4th of June is generally considered to be the genesis of punk rock6.
Read full article from Brewer's CAP Theorem <= :julianbrowne
Non-relational databases, also known as NoSQL databases, have emerged in recent years as an alternative to relational databases. Cloud Datastore is one of the most widely used non-relational databases in the industry. In 2013 Cloud Datastore processed 4.5 trillion transactions per month (Google Cloud Platform blog post). It provides a simplified way for developers to store and access data. The flexible schema maps naturally to object-oriented and scripting languages. Cloud Datastore also provides a number of features that relational databases are not optimally suited to provide, including high-performance at a very large scale and high-reliability.
Read full article from Balancing Strong and Eventual Consistency with Google Cloud Datastore | Cloud Datastore Documentation | Google Cloud Platform
某餐馆有n张桌子，每张桌子有一个参数：a 可容纳的最大人数； 有m批客人，每批客人有两个参数:b人数，c预计消费金额。 在不允许拼桌的情况下，请实现一个算法选择其中一部分客人，使得总预计消费金额最大。
输入包括m+2行。
第一行两个整数n(1 <= n <= 50000),m(1 <= m <= 50000)
第二行为n个参数a,即每个桌子可容纳的最大人数,以空格分隔,范围均在32位int范围内。
接下来m行，每行两个参数b,c。分别表示第i批客人的人数和预计消费金额,以空格分隔,范围均在32位int范围内。
采用贪心法。先把顾客进行消费金额降序，如果消费金额相等，则按照人数升序排序。 然后枚举每波顾客去二分当前最适合的桌子的容量。时间复杂度：O(mlogm + nlogm)
Read full article from 餐馆就餐问题 - ljp1919的专栏 - CSDN博客
How to Improve Your Skills as a Programmer: 11 Steps
Programming is one of the most versatile skills on the market in this age. From being able to create company websites to knowing how to easily fix a redirecting error, these skills can be invaluable to an employer and yourself in many ways. However, staying the way you are will never let you be the best programmer you can be. Read on to learn how to improve your skills as a programmer.
Read full article from How to Improve Your Skills as a Programmer: 11 Steps
Converting TimeUUID Strings to Dates – DataStax Support
TimeUUID is one of the datatypes supported by Cassandra. The Cassandra JDBC driver will return a hexadecimal representation of the UUID rather than a human-readable date. You can use the java.util.UUID and java.util.Date classes to convert from a TimeUUID string to a Date. The UUID.timestamp() returns a long containing the timestamp value associated with this UUID. The resulting timestamp is measured in 100-nanosecond units since midnight, October 15, 1582 UTC. Notice that this is a different epoch and resolution than the long expected by Date's constructor, which is the number of milliseconds since January 1, 1970, 00:00:00 GMT. Therefore, the conversion requires some math. Here is some example code to convert from a string representing a TimeUUID string to a java.util.Date object:Read full article from Converting TimeUUID Strings to Dates – DataStax Support
Converting TimeUUID Strings to Dates – DataStax Support
TimeUUID is one of the datatypes supported by Cassandra. The Cassandra JDBC driver will return a hexadecimal representation of the UUID rather than a human-readable date. You can use the java.util.UUID and java.util.Date classes to convert from a TimeUUID string to a Date. The UUID.timestamp() returns a long containing the timestamp value associated with this UUID. The resulting timestamp is measured in 100-nanosecond units since midnight, October 15, 1582 UTC. Notice that this is a different epoch and resolution than the long expected by Date's constructor, which is the number of milliseconds since January 1, 1970, 00:00:00 GMT. Therefore, the conversion requires some math. Here is some example code to convert from a string representing a TimeUUID string to a java.util.Date object:Read full article from Converting TimeUUID Strings to Dates – DataStax Support
Spring Boot - Cannot determine embedded database driver class for database type NONE - JournalDev
If you are reading this, then I think you have got "Cannot determine embedded database driver class for database type NONE" error while running Spring Boot application. Below image shows my simple Spring Boot starter project where I got "Cannot determine embedded database driver class for database type NONE" error while running the JUnit test case.Read full article from Spring Boot - Cannot determine embedded database driver class for database type NONE - JournalDev
When you just press "g" the browser receives the event and the entire auto-complete machinery kicks into high gear. Depending on your browser's algorithm and if you are in private/incognito mode or not various suggestions will be presented to you in the dropbox below the URL bar. Most of these algorithms prioritize results based on search history and bookmarks. You are going to type "google.com" so none of it matters, but a lot of code will run before you get there and the suggestions will be refined with each key press. It may even suggest "google.com" before you type it.
Read full article from GitHub - vasanthk/how-web-works: What happens behind the scenes when we type www.google.com in a browser?
A plain English introduction to CAP theorem « Kaushik Sathupadi
You'll often hear about the CAP theorem which specifies some kind of an upper limit when designing distributed systems. As with most of my other introduction tutorials, lets try understanding CAP by comparing it with a real world situation.
Last night when your spouse appreciated you on remembering her birthday and bringing her a gift, a strange Idea strikes you. People are so bad in remembering things. And you're sooo good at it. So why not start a venture that will put your talent to use? The more you think about it, the more you like it. In fact you even come up with a news paper ad which explains your idea
Read full article from A plain English introduction to CAP theorem « Kaushik Sathupadi
A Practical Approach to Computer Systems Design and Architecture - CodeProject
3. How do documents flow through this system?
*Select* and *upload* the document by the user > System will validate the input file > *Files added* to "to be approved" section > Administrator *Approves and publishes* the file > File is available for users.
3.1. Do we need a workflow management system here?
I don't think so, let's hardcode the workflow, since we don't expect them to be dynamically adjusted. This is not a very complex system. So it is not necessary to have a separate module to manage workflows, and I am happy without it.
Now, we have found a few granular level operations belonging to the document uploading section. Let's list them as below:
Client side:
Select a document from the local directory base
Read the byte stream to system memory
Send it via the web to the server
Server Side:
Read the byte stream from memory
Write it to a local temporary location for validations
Input Validation Module validates the input file
Move the validated file to the correct user's shared location
Add this new file to the repository system as a "File to be Approved"
Load the files that are to be approved for the system administrator to view them
System administrator validates the content of the file and approves them to be viewed online
Change the file metadata status from "tobeApproved" to "Approved"
Note: Ideal place for a diagram..
3.2. Are we going to have versioning here?
Yes, we need to have versioning. So any update to an existing file will run through a version handler that assigns the correct version number. Note: This can be analyzed further deep..
3.3. Do we need to notify users when a new document is successfully published?
The user who owns (uploads) the document will get an email noting the reject/ accept status. In addition to that, we have to send another notification to users who have subscribed to be notified as this operation happens. Note: This introduces a new module to the system named Notification module. I will not dig into this section either.
3.3. Can a user undo an update?
I need to talk to the client about this.
3.4. Can multiple users checkout a document at the same time?
Let's stop this now…
Read full article from A Practical Approach to Computer Systems Design and Architecture - CodeProject
When you just press "g" the browser receives the event and the entire auto-complete machinery kicks into high gear. Depending on your browser's algorithm and if you are in private/incognito mode or not various suggestions will be presented to you in the dropbox below the URL bar. Most of these algorithms prioritize results based on search history and bookmarks. You are going to type "google.com" so none of it matters, but a lot of code will run before you get there and the suggestions will be refined with each key press. It may even suggest "google.com" before you type it.
Read full article from GitHub - vasanthk/how-web-works: What happens behind the scenes when we type www.google.com in a browser?
Stateful Streaming in Spark and Kafka Streams
TLDR: This article is about aggregates in stateful stream processing. It covers two concrete examples in Apache Spark (using the Streaming API with mapWithState) and Apache Kafka (using the high-level DSL in Streams). Spark Streaming and Kafka Streams differ much. Therefore the reader can get to know both approaches and can decide which fits best. I discuss what features for real-time aggregates besides reliability and scaling requirements we would like to see covered (e.g. to be queryable, re-processable, versionable, composable, unlimited updates, providing retention times, downsampling and having end-to-end guarantees). In the end, I will discuss the example code, also along those requirements.
Read full article from Stateful Streaming in Spark and Kafka Streams
Introducing Kafka Streams: Stream Processing Made Simple - Confluent
I'm really excited to announce a major new feature in Apache KafkaTM v0.10: Kafka's Streams API. The Streams API, available as a Java library that is part of the official Kafka project, is the easiest way to write mission-critical real-time applications and microservices with all the benefits of Kafka's server-side cluster technology.
Read full article from Introducing Kafka Streams: Stream Processing Made Simple - Confluent
Capture Screenshot - Chrome DevTools - Dev Tips
You can type in screenshot into the Command Menu (shortcut: Cmd + Shift + P) and select one of two options:
The "Capture screenshot" option will download an image of your website based on what is in the visible viewport.
Read full article from Capture Screenshot - Chrome DevTools - Dev Tips
Capture Screenshot in Chrome without any tool / plugin / extension – Abhishek Shukla
All of us need to capture the screenshot of the browser one time or another and I have used many third party freemium extensions over time but nothing could be as good as something built in to the browser itself.
Chrome now includes a way to capture the screen shot through Dev Tools. To Open Dev Tools Press F12 or Ctrl + Shift + I or from the hamburger menu in Chrome select More Tools –> Developer Tools
Read full article from Capture Screenshot in Chrome without any tool / plugin / extension – Abhishek Shukla
Principles are general rules and guidelines, intended to be enduring and seldom amended, that inform and support the way in which an organization sets about fulfilling its mission.
In their turn, principles may be just one element in a structured set of ideas that collectively define and guide the organization, from values through to actions and results.
Depending on the organization, principles may be established at any or all of three levels:
These sets of principles form a hierarchy, in that IT principles will be informed by, and elaborate on, the principles at the enterprise level; and architectural principles will likewise be informed by the principles at the two higher levels.
The remainder of this subsection deals exclusively with architectural principles.
Read full article from Architectural Principles
Diverse Teams Feel Less Comfortable — and That's Why They Perform Better
In numerous studies, diversity — both inherent (e.g., race, gender) and acquired (experience, cultural background) — is associated with business success. For example, a 2009 analysis of 506 companies found that firms with more racial or gender diversity had more sales revenue, more customers, and greater profits. A 2016 analysis of more than 20,000 firms in 91 countries found that companies with more female executives were more profitable. In a 2011 study management teams exhibiting a wider range of educational and work backgrounds produced more-innovative products. These are mere correlations, but laboratory experiments have also shown the direct effect of diversity on team performance. In a 2006 study of mock juries, for example, when black people were added to the jury, white jurors processed the case facts more carefully and deliberated more effectively.
Read full article from Diverse Teams Feel Less Comfortable — and That's Why They Perform Better
Diverse Teams Feel Less Comfortable — and That's Why They Perform Better
In numerous studies, diversity — both inherent (e.g., race, gender) and acquired (experience, cultural background) — is associated with business success. For example, a 2009 analysis of 506 companies found that firms with more racial or gender diversity had more sales revenue, more customers, and greater profits. A 2016 analysis of more than 20,000 firms in 91 countries found that companies with more female executives were more profitable. In a 2011 study management teams exhibiting a wider range of educational and work backgrounds produced more-innovative products. These are mere correlations, but laboratory experiments have also shown the direct effect of diversity on team performance. In a 2006 study of mock juries, for example, when black people were added to the jury, white jurors processed the case facts more carefully and deliberated more effectively.
Read full article from Diverse Teams Feel Less Comfortable — and That's Why They Perform Better
Read full article from line-ending-converter
The Seven Deadly Sins of Solr | Lucidworks
Read full article from The Seven Deadly Sins of Solr | Lucidworks
Solr Multithreaded concurrent atomic updates
Solr Multithreaded concurrent atomic updates problem:
Solr has few limitations for the data ingestion, as it doesn't provide row level lock over document.
I face this problem while uploading data in bulk to solr5 in multithread environment and I solved it by solrj client side lock.
When concurrent threads try to make atomic update on a multivalued field of a document at the same time, few threads changes get overridden and it happens because last thread update take sometime to get indexed.
Data ingestion scenario:
There are two tables in RDBMS and I need to denormalize in solr, Steps I was following for atomic/partial document update-
1- Fetch the existing document.
2- Update the single value fields if required and add/set the new values to multivlaued fields.
3- Update the final document back to solr.
Read full article from Solr Multithreaded concurrent atomic updates
Most often, the term is used for those situations when this is not the intended result; that is, when this is a bug. Such errors are most common among novice programmers, but can be made by experienced programmers as well, because their causes can be quite subtle.
One common cause, for example, is that the programmer intends to iterate over a collection of items such as a linked list, executing the loop code once for each item. Improperly formed links can create a reference loop in the list, where one list element links to one that occurred earlier in the list. This joins part of the list into a circle, causing the program to loop forever.
While most infinite loops can be found by close inspection of the code, there is no general method to determine whether a given program will ever halt or will run forever; this is the undecidability of the halting problem.
Read full article from Infinite loop - Wikipedia
Writing Bug-Free C Code: General Tips
Once the create and destroy methods are working correctly, it is time to start implementing the other method functions. The best strategy is to implement a method function, compile it and then test it by writing test code.Read full article from Writing Bug-Free C Code: General Tips
Deepak Vadgama blog – Spring Boot Wonders
Spring Boot is mighty powerful. It enables multitude of functionalities with minimal amount of code (thanks to annotation processing & sensible defaults).
Listed below are a few features I've been using in a new project. I am creating a pre-configured scaffold code repository, will share the same soon.
Read full article from Deepak Vadgama blog – Spring Boot Wonders
Deepak Vadgama blog – Java Interview Study Notes
I created some study notes for my interviews in early 2017. I've posted them to Github so that others can benefit from it. These notes are specifically for theory based Java interviews, typically done in service based companies (as opposed to coding based interviews done in product companies). Hope it helps.
Read full article from Deepak Vadgama blog – Java Interview Study Notes
Thread Magic Tricks: 5 Things You Never Knew You Can Do with Java Threads | Takipi Blog
Some people like mountain climbing, others do sky diving. Me, I like Java. One of the things I love about it is that you never stop learning. The tools you use on a daily basis can often reveal a whole new side to them, with methods and interesting use cases you haven't had a chance to see yet. Like threads for example. Actual threads. Or better put, the Thread class itself. Concurrent programming never stops posing challenges when we're dealing with high scalability systems, but now we'll talk about something a bit different.
In this post you'll see some of the lesser known yet useful techniques and methods that threads support. Whether you're a beginner, advanced user or an expert Java developer, try to see which of these you already know and what comes off as new to you. Is there something else about threads you feel worth mentioning? I'd love to hear about it in the comments below. Let's get started.
Read full article from Thread Magic Tricks: 5 Things You Never Knew You Can Do with Java Threads | Takipi Blog
决策树(decision tree)是一种基本的分类与回归方法。本章主要讨论分类的决策树。
定义 5.1(决策树) 分类决策树模型是一种描述对实例进行分类的树形结构。决策树由结点(node)和有向边(directed edge)组成。结点有两种类型：内部结点(internal node)和叶结点(leaf node)。内部结点表示一个特征或属性，叶结点表示一个类。
用决策树分类，从根结点开始，对实例的某一特征进行测试，根据测试结果，将实例分配到其子结点；这时，每一个子结点对应着该特征的一个取值。如此递归地对实例进行测试并分配，直至达到叶结点。最后将实例分到叶结点的类中。
Read full article from 第5章 决策树 | 遇见一座城
About me - Federico Tomassetti - Software Architect
I believe that software is a tool which improve so many of our activities: by providing software we can open so many possibilities.
People have unique skills and competencies and software should be there to complement them, to help them performin analysis quickly, to help them visualize ideas, to run checks and calculations for them. So that people can focus on using their abilities to provide more value.
To me building software means to provide tools to people.
Read full article from About me - Federico Tomassetti - Software Architect
Google Code Archive - Long-term storage for Google Code Project Hosting.
Hamcrest has been designed from the outset to integrate with different unit testing frameworks. For example, Hamcrest can be used with JUnit 3 and 4 and TestNG. (For details have a look at the examples that come with the full Hamcrest distribution.) It is easy enough to migrate to using Hamcrest-style assertions in an existing test suite, since other assertion styles can co-exist with Hamcrest's.
Hamcrest can also be used with mock objects frameworks by using adaptors to bridge from the mock objects framework's concept of a matcher to a Hamcrest matcher. For example, JMock 1's constraints are Hamcrest's matchers. Hamcrest provides a JMock 1 adaptor to allow you to use Hamcrest matchers in your JMock 1 tests. JMock 2 doesn't need such an adaptor layer since it is designed to use Hamcrest as its matching library. Hamcrest also provides adaptors for EasyMock 2. Again, see the Hamcrest examples for more details.
Read full article from Google Code Archive - Long-term storage for Google Code Project Hosting.
这个原则是今日头条刷掉我的简历的时候给我的回复，他们希望我按照这个原则再好好修改一下简历。在这里谢谢他们！
HR会更看重工作能力。
如果有以下问题，可以视为不及格。
如果有以下问题，可以视为及格，但还不够好。
Read full article from 如何写好简历 - 墨染百城 - CSDN博客
手写识别系统（k-近邻算法） - marlonlyh的博客 - CSDN博客
k-近邻算法（kNN）是机器学习中一个相对比较简单的算法。该算法在训练集中数据和标签已知的情况下，输入测试数据，将测试数据的特征与训练集中对应的特征进行相互比较（比如通过欧氏距离），找到训练集中与之最为相似的前K个数据，则该测试数据对应的类别就是K个数据中出现次数最多的那个分类，其算法的描述为：
1）计算测试数据与各个训练数据之间的距离；
2）按照距离的递增关系进行排序；
3）选取距离最小的K个点；
4）确定前K个点所在类别的出现频率；
5）返回前K个点中出现频率最高的类别作为测试数据的预测分类。
以下代码是通过kNN算法实现的手写识别系统。其中每个输入数据为32*32的二值图像，数据存储在txt文件中（32行*32列）。文件命名为诸如"0_0.txt"的形式。其中，"_"左边的0为其实际标签（实际类），右边的0为数据的编号。
Read full article from 手写识别系统（k-近邻算法） - marlonlyh的博客 - CSDN博客
Uber is the most valuable U.S. startup, with Airbnb and WeWork following far behind it - Recode
Benchmark and SoftBank have been publicly negotiating an Uber deal that could let Benchmark pocket some of the cash it has made from Uber, and allow SoftBank to invest in the ride-hailing giant.
The conversation has led to a discussion of Uber's valuation. Uber investor Benchmark foresees it going as high as $100 billion — though, of course, it would have incentive to say that.
Read full article from Uber is the most valuable U.S. startup, with Airbnb and WeWork following far behind it - Recode
Find all the words in a list that are next to a given word | shanhe.me
There is a list of four-letter words, such as aahs, aals, abas, abba, etc. Given an arbitrary four-letter word, find all the words in the list that are next to the given word, i.e., words that can be the same as the given word with exactly one letter changed. Example: given puma, the result could be duma, pima, puja, pula, pump, puna, pupa.
We use an index-based solution, and we give the result containing unique and sorted words from the list.
Suppose we look up words for puma, then we
?uma with ? less than p.p?ma with ? less than u.pu?a with ? less than m.pum? with ? less than a.pum? with ? greater than a.pu?a with ? greater than m.p?ma with ? greater than u.?uma with ? greater than p.and combine the results in the order of the steps to get the final result. Note that if the result from each step contains unique and sorted words, then the final list contains unique and sorted words.
The index for unique and sorted words from the list matching a pattern like ?uma with ? less than or greater than p can be designed like
Read full article from Find all the words in a list that are next to a given word | shanhe.me
ThinkLeet/（柱状图型）2.1.15 Trapping Rain Water at master · YanjieGao/ThinkLeet · GitHub
Given n non-negative integers representing an elevation map where the width of each bar is 1, compute how much water it is able to trap after raining. For example, Given [0,1,0,2,1,0,1,3,2,1,2,1], return 6. 图2-3 Trapping Rain Water 解法：左遍历，右遍历，求出每个柱子，最上方能称多少水。最后累加。时间复杂度o(n), 空间复杂度o(n) 变式：为为2D的模型来做。围住的空间是多大。 二维的解法： 方法1：类似trap water 方法2：可以反证法，求出x的空间然后总空间减去这个空间 三维也是同理，能够围住多少水 变式： （1）柱子内的最大长方形面积(4.1.3 Largest Rectangle in Histogram)：栈做，时间复杂度o(n)，空间复杂度o(n) (2)两个柱子组成的最大长方形面积：夹逼法，小的向里移动，（因为大的向内移动只能减少面积，小的还有希望扩展面积）。 时间复杂度o(n)空间法复杂度o(1) （3）生成新柱状图型：原始是权重柱子，新的柱状图每个节点如果权重大于neighbour则新的柱状图中也要值大于neighbor（2.1.22 Candy） 方法：左遍历,右遍历。时间复杂度o(n)，空间复杂度o(1) 变式：权重大的反而要小，只要把之前算法的权重改为-1值即可，算法仍可复用。 (4)Stock模型： 1 一天买卖一次：(12.3 Best Time to Buy and Sell Stock) 遍历，统计min和统计maxprofit，类似动归的思想 2 一天买卖n次：(12.4 Best Time to Buy and Sell Stock II) 差分数组。 3 一天买卖2次：(13.5 Best Time to Buy and Sell Stock III) 动归，分两种情况 第一类 两个事务相交 0 i n ，在0 i 和i n分别存在两个事务，然后每个里转化为问题1. 第二类 两个事务不相交 0 i n两个最小值在 0 i两个最大值在i n，从左遍历一次统计所在节点左侧的两个min ，从右遍历一次统计右侧的两个max，这个过程 就可以求出i点的maxprofict 时间复杂度 o(n2)空间复杂度 o(n) 4 变式：一天买卖k次：类似3一天买卖2次的处理方式.相比1,2要复杂 还是分类 1个和k-1个事务不相交： 0 i n，转换为子问题 f(1)和f(k) 1个和k-1个事务相交：分类和其中 1个，2个，3个。。。k个相交做。 （5）并行化（一般从头遍历，切求最值的不太好并行化）： 2.1.15 启示：可以思考多轮次做，前几轮进行元数据信息获取，或者进行sample 进行range分区，每个分区的柱子还是原顺序。 第一轮：获取这个分区的max高度。map为key（左右坐标），value（max高度），reduce更新各分区的左右做大高度。 第二轮：然后按照单机的做法求最大的水量Read full article from ThinkLeet/（柱状图型）2.1.15 Trapping Rain Water at master · YanjieGao/ThinkLeet · GitHub
Intellisense-like auto-complete – HackerRank Support Center
HackerRank's editor has an auto-complete feature similar to what is found in modern editors such as Eclipse or Visual Studio. The auto-complete automatically gets triggered for languages such as Java and C# at the specific syntax points.
Example: When you type a full stop (.) in System.out.
You can also trigger it manually at any time by using the combination of CTRL and Space key.
Read full article from Intellisense-like auto-complete – HackerRank Support Center
挑战程序竞赛系列（19）：3.1最小化第k大的值 - Demon-初来驾到 - CSDN博客
针对区间的二分头一次见，不过很有意思，总的思想是，给定了一个mid值之后，统计左区间符合total的个数left和右区间的个数right，如果left较小，说明mid不够大，更新lb，而左右区间个数均符合的情况下，更新lb，因为总是求较大的mid。
Read full article from 挑战程序竞赛系列（19）：3.1最小化第k大的值 - Demon-初来驾到 - CSDN博客
硅谷IT公司怎么操作裁员？拿Twitter做例子说说 | 美国留学就业规划|Warald咨询
那这次裁掉的都是哪些人呢？看看思路，大家可以借鉴。
1. 不来钱、没有钱景的部门。
养着这样的队伍对公司来说是烧钱。上至领导下至小兵，全砍掉，即使你是浴血奋战、武艺高强的小兵，既然战局不利，就随着将军阵亡吧。
2. Open source project
很多人都说，在公司里做open source技术最好了，将来跳槽去外面机会多啊，很多公司都用这些技术。如果用的是某家公司的内部开发工具，而下家公司又不用，就没优势了。很不幸，脱离了盈利目标的技术流，在公司动荡期很难自保，被twitter现在就送到外面去了~
据Warald所知，twitter做open source的组很强，员工很优秀。问题是这些项目侧重长期影响力，短期内不会给公司带来直接利润，作为上市公司，在财政压力下，长期的项目只能放弃。
3. 各个组工作表现垫底的人（low performer）
裁员之际，每个组里表现差的人，只能出局，这个大家都能理解。不过，我想做一些说明：Twitter是一线公司，人才储备丰富，一流人才很多，但是据Warald所知，很多组里风气不好，老人压新人或者抢credits（把别人的活据为己有、抢别人功劳）这种事情，时有发生，所以即使是low performer，只要是能在公司里能存活的，未必就是能力不足。换个地方，也许有新机会吧。
前几年就业形势好，很多人春风得意觉得高薪工作、优厚待遇是理所当然的。在拼命招人的时候，公司当然要说工程师们是我们的宝贵财富，给高薪，许诺大把的期权股票，Twitter就是这方面的典范 – 注意，Warald不是以讽刺的口吻来说的。实际上，Twitter以前对员工的确非常好，即使是公司上市之前不久加盟的员工，也有很多人拿到上百万的股票，在硅谷公司里算是首屈一指的。
Read full article from 硅谷IT公司怎么操作裁员？拿Twitter做例子说说 | 美国留学就业规划|Warald咨询
【新提醒】Snapchat vs Facebook vs Google vs Airbnb vs Uber vs Linkedin【一亩三分地论坛找工求职版】 - Powered by Discuz!
FB是我目前最喜欢的一家，他家现在growth确实很强劲，里面机会很多，另外品牌效应也最强，对将来的发展会很好。虽然要经过bootcamp，但是如之前所说，他家现在各个领域都在扩张，其实还蛮容易找到喜欢的组的。如果要去FB家的话，一个建议就是LZ可以考虑选较新的组，因为这种组更容易出成绩，而且更容易随着组的发展做核心。比较老的组会有很多资历和等级在你上面的人，你有可能找不到足够有credit的项目去升级，学习东西。还有一点，FB他家的code base非常的好，技术真的是狗家和他家是业界最强的，进去可以学到很多东西。缺点就是会很累。最后的一个优点，FB和狗家的文化一样，都是tech driven，是engineer有最大的话语权，这一点很多新人和没工作经验的人体会不到，但是只要你工作个几年，应该都会有很深的体会。这个部门或者群体在公司有没有很大的话语权真的很不一样的。Read full article from 【新提醒】Snapchat vs Facebook vs Google vs Airbnb vs Uber vs Linkedin【一亩三分地论坛找工求职版】 - Powered by Discuz!
The Right Way to Follow Up After a Job Interview
Thank you so much for meeting with me today. After learning more about the position, I'm very excited about the opportunity to join your team and help [create world-class marketing campaigns, inspire prospective clients, increase revenue, etc.] for [company name].
Read full article from The Right Way to Follow Up After a Job Interview
LeetCode 638. Shopping Offers · Shell32
In LeetCode Store, there are some kinds of items to sell. Each item has a price.
However, there are some special offers, and a special offer consists of one or more different kinds of items with a sale price.
You are given the each item's price, a set of special offers, and the number we need to buy for each item. The job is to output the lowest price you have to pay for exactly certain items as given, where you could make optimal use of the special offers.
Each special offer is represented in the form of an array, the last number represents the price you need to pay for this special offer, other numbers represents how many specific items you could get if you buy this offer.
Read full article from LeetCode 638. Shopping Offers · Shell32
[心得] 第二次 Google 面試洗臉心得 at Soft_Job - BWall 熱門 ptt bbs 文章
老實說得知結果的當下真的有點沮喪，各種負面情緒混著出現 （「花了一整年卻是這種結果」、「四種結果還被解讀成是狀況不穩」之類的）， 不過冷靜下來反省一下面試當時的狀況，也確實有不少應該改進的地方： 1. 說演算不行的可能是第一位考 Fenwick Tree 的 這只能怪自己當時沒有充分準備、還有沒把握解釋就選擇繞路的反應吧 2. 只有第二位和最後一位有考設計題 不過感覺上應該是最後 LRU Cache 裡資料結構用成 Heap 問題比較大 3. 這樣看來，說 coding 不行的應該就是希望我找中位數的第二位了 所以就當成是一個寶貴的經驗，下次再加油吧。 ……當時我是這麼想的。 兩個月後我在一次偶然機會下拿到了 wish list 裡另一家公司的 offer 雖然還是有朋友希望我繼續挑戰 Google 不過老實說自己已經不太想繼續過這種只有工作 + 準備面試的生活 （相對犧牲了運動、學其他東西或是搞自己 project 的時間，生活品質也不好） 所以最後選擇拿了那份 offer ，現在正在收行李準備展開新的生活 至於這次經驗就分享給有興趣挑戰的板友，希望能有所幫助。 祝有志之士能塊陶成功！ 追記： 沒想到這篇會吸到將近 50 推和一堆恭喜（一堆恭喜的洗臉文……開玩笑的） 後面會補些板友們比較好奇的內容 1. 為什麼不在臺北面試？ 主要是因為 recruiter 在安排 on-site 前就有和我說明臺北目前人手嚴重不足 若要安排在臺北面試，面試官的安排可能相當難處理 必要場合會請其他 branch 的面試官透過視訊面試或是拆成兩天進行 （若面試者第一天表現不佳，甚至有可能直接取消其他天的面試） 由於我是在職應徵，面試需要請假（而且是同事一看就知道有鬼的那種請法） 所以自然不希望 recruiter 把面試拆成太多天進行Read full article from [心得] 第二次 Google 面試洗臉心得 at Soft_Job - BWall 熱門 ptt bbs 文章
[心得] 第二次 Google 面試洗臉心得 at Soft_Job - BWall 熱門 ptt bbs 文章
標題寫得非常清楚，這篇是我去年第二次挑戰應徵日本 Google …… 然後再次被洗臉的故事。 第一次在臺北的面試由於 NDA 的關係不能分享整體面試內容 但是這次日本面試有特別確認過不用簽 NDA ，所以下面會講得很詳細 當然這不保證之後會不會被查水錶就是了，要是被查會回來把這篇毀屍滅跡 聯絡 recruiter （應該說被聯絡）： 這次的面試並不是我主動投的，是 Google 一位負責亞太地區的 recruiter 找上門來： 「我有看到你去年應徵的記錄，請問你是否還有興趣再次應徵 Google 呢？」 其實我本來是打算在年底準備得差不多時自行應徵（因為我原先也是規劃一年準備期） 不過當時可能是想說這次居然有 recruiter 主動找上門來，擇日不如撞日 於是就馬上把更新好的履歷寄回去，開啟了第二次（洗臉）挑戰 面試準備： 除了履歷以外，其他項目都是在第一次面試失敗後開始利用下班時間準備的 所以這樣算起來，整個準備期間大概是十個月左右 如果是辭掉工作全心準備的話，聽其他有嘗試過的朋友提大概都是抓三個月 履歷： Google 面試時對於履歷內容的提問比例相對於其他公司算是比較少的 不過也不能保證完全沒有，所以面試前還是必須仔細確認自己寫在履歷裡的內容 確定到時候萬一真的被問起不會直接愣在原地 程式語言：由於四年工作經驗全是在做 Android 開發，我自然選擇了用 Java 應試 不過這裡的重點是一旦選定要用的程式語言就必須要詳細複習那個語言的語法 因為有些對程式碼特別要求的面試官會拿面試者的程式碼追問「為什麼要這樣寫」 （「為什麼 static 」「為什麼不用 float 」「這 call by reference 安全嗎」等） 所以比起「知道很多語言」，「把一種語言用到專精」反而比較重要 Leetcode ：近幾年 Leetcode 的題目數量膨脹得相當快 直到我 on-site 面試前已經有大約 400 題，我自己是陸陸續續做了大概 340 題 如果是要挑戰 Google 的話，大致上有幾個利用方向： 1. 訂閱：這不是幫 Leetcode 打廣告（強調） 主要是我自己跟幾個有挑戰的朋友面試時都有碰到被 Leetcode 鎖住的題目 所以即使沒有自己付錢訂閱，有機會的話一定要打聽看看鎖住的題目 2. 參加比賽： Leetcode 現在每週末都有辦比賽，比賽中會放出新的題目 運氣好的話會碰到之後會鎖起來的題目，等於不用訂閱就看到題目了 3. 如果時間有限，練好一定基礎之後就從最後一題往前做 因為越後面的題目通常都是最近去面試的挑戰者回報的新題目 4. 最後就是碰碰運氣了，因為最近回報的題目裡出現賽局或是證 NP-Hard 的題目…… 基本上準備面試時不太可能涵蓋所有類型的題目，所以這種的就只能準備多少算多少了 Code Jam ：由 Google 舉辦的程式競賽，官方會在比賽結束後放出題目和解答 有些面試官會拿比賽題目當面試題目 系統設計：就是那種「要你設計一個 Twitter 或是設計一個 Amazon 」的題目 這種題目對於沒有雲端背景的面試者來說確實比較難以下手 但是很不幸的，不只 Google ， FLAG 或同等級的大型網路公司面試都很常出這種問題 （有傳聞說掛五年以上資歷的工程師面試 Google 一定會遇到至少一題系統設計題） 我自己也沒有相關經驗，只能在面試前到 HiredInTech 網站上看看要怎麼應對這種題目 那個網站上面很詳細地解釋了該怎麼一步一步地進行系統設計討論 （釐清需求→設計架構→發現瓶頸→改進架構），對於沒經驗的面試者幫助很大 電話面試： 我當時的打算是在電話面試上速戰速決（馬上安排面試，不再花時間準備） 這樣萬一電話面試就翻船的話，我就可以馬上評估要不要準備其他公司的面試 所以其實我這次有在電話面試上小賭了一把，賭電話面試只會出 medium 以下的難度 （聽說運氣很背的話電話面試就會被 hard 級題目洗臉） 這樣以我八月當時的準備狀況應該就足以應付大部分的題目 結果我運氣還不錯，這次碰到經典題大數相加（兩個超出 long 長度的數字相加） 題目本身不難，但是卻碰到一位對語言相當要求的中國面試官 我在程式碼中為了保險起見加的 explicit casting 全被他抓出來問是否真的有必要加 （他本人沒有表達預設立場，純粹是要測試我對 Java 的 casting 規則熟不熟） String 和 StringBuilder （ mutable string ）的使用也被提出來討論 幸好最後還是得到面試官認可，安全進入下一關 On-site ： 電話面試完四天後，我收到另一位 recruiter 來信通知準備安排 on-site 面試 （聽說第一位聯絡我的 recruiter 只負責到電話面試結束，所以後面交棒了） 不過當時我提了很多麻煩的條件： 1. 由於當時我還需要一些時間準備，所以希望把 on-site 時間排到兩個月後 2. 我跟 recruiter 提出希望能「自費飛去日本的 Google Office 面試」 原因還滿多的，包括臺北人力吃緊，排面試的時間限制非常多（很可能要分兩天） 以及希望能先去看看日本辦公室的樣子、還有同學建議不要去臺北 office 面試之類的 後來喬了好一會兒才終於敲定十月中到日本面試 當天的面試官一共四位，沒有實習面試官；全程以英文（我選的語言）進行 第一位美藉面試官考了 Leetcode #308 ，但是要把取總和的函式改成取平均 當時我一聽到這題心就涼了一半，因為我知道這題會扯到 Fenwick Tree 但是整個準備過程中用到 Fenwick 的題目不到 1% ，所以我沒有練熟…… （原理大致上能解釋，但是沒把握能寫成 code ） 或許是出於保守心態吧，最後我選擇忽略 Fenwick Tree 向面試官提出 naive 和 prefix sum 兩種解法、然後照要求 coding 和比較複雜度 另外由於這題被改成取平均，所以有討論到要以什麼型態回傳（精度）的問題 第二位面試官是個嚴肅的日籍中年工程師，看起來就像一般日本上班族： 1-1. 什麼是中位數？什麼狀況下會考慮使用中位數而非平均數？ 1-2. 寫一個函式吃三個數字吐其中的中位數、並附上 unit test 測資 1-3. 你的函式執行一次最多會用到幾個 > 、 < 或 = ？有沒有辦法減少？ 為了涵蓋到所有 case ，我在寫函式時直接窮舉了所有可能組合， 結果最多會比到五次，寫完後馬上被面試官用 1-3 （寫完後才問）打臉 最後印象中可以砍到最多只要比三次 2. 設計一個負責收集 log 的系統；這個系統要能收集從多個伺服器傳來的 log 、 並讓工程師調出指定時段內的所有 log 第三位面試官是位年輕的日本工程師，面試經驗比較淺： 1. 看履歷上摸過什麼技術、然後問該技術相關的東西（例如 Unicode ） 2. Road to Freedom （遊戲 Fallout 4 裡的任務） 首先會播一段該任務的攻略 https://youtu.be/v6_cRSULPZ8 （有玩的板友注意有雷）Read full article from [心得] 第二次 Google 面試洗臉心得 at Soft_Job - BWall 熱門 ptt bbs 文章
LeetCode Weekly Contest 22 | 小小小宇
Given an array of integers and an integer k, you need to find the number of unique k-diff pairs in the array. Here a k-diff pair is defined as an integer pair (i, j), where i and j are both numbers in the array and their absolute difference is k.
Example 1:
|
|
Example 2:
|
|
Example 3:
|
|
Note:
Read full article from LeetCode Weekly Contest 22 | 小小小宇
LeeCode Combination Sum 更新动态规划法 - 靖空间 - CSDN博客
Combination Sum
Given a set of candidate numbers (C) and a target number (T), find all unique combinations in C where the candidate numbers sums to T.
The same repeated number may be chosen from C unlimited number of times.
Note:
For example, given candidate set 2,3,6,7 and target 7,
A solution set is:
[7]
[2, 2, 3]
Because it is not confined to a specific length of the answers, so we cannot kown that before testing, which means there are not fixed end point of the backtracking.
We have to prun(剪枝） the branch(分支）, and backtrack to the last number.
It work like this:
1 We sort the numbers in non-descendant way
排序
2 Add the smallest number first, and loop to add it again, and again, until we find a solution, or the solution's value is larger than our target number.
迭代或递归求解
3 We backtrack, by pop up the number , to the last number.
回溯
4 and then we try the next number in candidates, loop to 2.
The key thoughts to modle the solution are:
1 we use pushing number in and popping out to stand for backtracking thought.
回溯思想由push和pop体现
2 How many numbers in our solution stand for how deep level we iterate or recur.
3 we stop further iterate or recur by compare the number value in solution to the target number.
4 and further more, by knowing the way we arrange our candidates in a non-desendant way, we know if the current number won't fit for the answer, that the rest of the number in candidates won't fit too, so we don't need to try the rest. That's also the thought popular in AI, which call Heuristic Searching.
经验剪枝法
How we deal with repeated answers?
We just forward trying, don't look back the numbers in candiates.
At first, I was wondering whether we can do it in a iterative(迭代循环）way.
So I try, and it work. Below is my iterative program:
Read full article from LeeCode Combination Sum 更新动态规划法 - 靖空间 - CSDN博客
Why is Kafka so fast | Searene
As we know that Kafka is very fast, much faster than most of its competitors. So what's the reason here?
Kafka writes everything into disk in order and consumers fetch data in order too. So disk access always works sequentially instead of randomly. Due to characteristics of disks, sequential access is much faster than random access. Here is a comparison:
Read full article from Why is Kafka so fast | Searene
39. Combination Sum
Given a set of candidate numbers (C) and a target number (T), find all unique combinations in C where the candidate numbers sums to T.
The same repeated number may be chosen from C unlimited number of times.
Note:
For example, given candidate set [2, 3, 6, 7] and target 7,
A solution set is:
[ [7], [2, 2, 3] ]
思路：
可以重复使用元素。那么每次调用从当前位置开始，如果当前位置元素小于目标元素，加入path，再次从当前位置调用；每个call中都要遍历后面的可能位置一遍。因此排序+剪枝。
Read full article from Leetcode 39 Combination Sum & 40 Combination Sum II & 216 Combination Sum III & 377 Combination V - fantasia的博客 - CSDN博客
LeetCode 446. Arithmetic Slices II - Subsequence - Mr.Phoebe的专栏 - CSDN博客
一道长得一副dp的样子的dp题。
这道题难度不算特别大，因为看得出来肯定是dp题。因为，一个等差序列里面有好几个小的等差序列。
例如，2 4是一个等差序列，2 4 6是一个等差序列。
所以我们发现等差序列是可以扩展的。
那么就得到了，咱们的转移方程的一部分
Read full article from LeetCode 446. Arithmetic Slices II - Subsequence - Mr.Phoebe的专栏 - CSDN博客
Maximal Rectangle: [LeetCode] 576. Out of Boundary Paths
Read full article from Maximal Rectangle: [LeetCode] 576. Out of Boundary Paths
借助软件养成新习惯不容易，我们试了上百应用只有这 8 个坚持超过半年 | Hack Your Life – CodingBlog
你的生活中可能会面临很多自我对抗。
想要减肥塑形，也明白要控制饮食，但还是忍不住在工作日的下午吃零食；知道早睡早起才会提高工作和生活效率，但晚上睡前还是止不住盯着屏幕一遍遍刷新信息。
习惯的力量可能比你想象中还要大，不管是好习惯还是坏习惯。
《纽约时报》记者查尔斯·杜希格所著的《习惯的力量: 为什么我们这样生活工作？》有一些可以参考的观点，它们解释了习惯到底是怎么一回事。
按照这本书的说法，习惯有 3 个部分，共同组成一个"习惯回路"（habit loop）——动机（cue）、惯常行为（routine）和奖励（reward）。动机让你展开行动、惯常行为就是习惯的动作、奖励促使你继续完成这个动作。在这三个部分的作用下，习惯是自动执行的，就像机器那样。
如果还是不太明白，可以回想一下日常生活中都有哪些固定的动作，来识别习惯的存在。比如"我想吃个冰淇淋"，你以为这是一种决策，其实它在某些情况下是习惯：下午 3 点多，工作效率到了一天最低的时候，你处在一种无聊的状态。动机产生，吃个甜筒获取满足感的奖励促使你每天下午都去买一个。
一样的道理，像是健身、读书这些健康的习惯，最好的状态是别人觉得你好厉害，但你自己并没有觉得有多难。因为它已经变成了一种惯常行为。
习惯的养成也可以从这三个方面入手，给自己一个动机，一定要简单易行切合实际；重复的动作，一次次强化对于动机的"渴望"，以及对奖励的"渴望"。
当然坚持重复动作也是最难的部分了，有时候形成某个互相督促和激励的小团体，甚至从个人信仰上获得力量也都是有点作用的。
到底多久才能养成一个习惯？你可能听过"21 天理论"的传说，但最近的研究表明，这的确也就只是一个传说。
"21 天养成一个习惯"一度被认为出自美国心理学家威廉・詹姆斯（William James），但其实没有人找到过他本人说这句话的出处所在。这中间可能有误解，整容医生麦斯威尔•莫尔茨（Maxwell Maltz）有另一个理论，整容病人平均需 要 21 天来习惯他们的新脸孔、截肢患者初期的"幻肢"体验往往需要 21 天才会消失。这和"习惯养成"没有什么关系。
所以不必纠结多久能养成一个习惯了，这没什么规律可循。伦敦大学学院的健康心理学家费莉帕・勒理（Phillippa Lally）的一项研究表明，志愿者习惯自动化的时长不一，有人 18 天就养成习惯，有人到 84 天实验结束都没有成功。这也和习惯本身的难易程度有关。
难度越低，奖励又诱人，坏习惯也就这样形成了，比如我们前面说到冰淇淋的例子。心理医生 Judson Brewer 在 TED 上做过一场演讲，他的核心观点是，革除坏习惯的动机中，如果我们不去跟大脑抗争、强迫自己专注，而是将它变成一种学习为本的过程，这个坏习惯会更容易被克服。
他用了戒烟的例子，强忍着让自己不去抽烟，这个痛苦的过程大多数人都无法坚持。他在试验中改用了冥想的办法，把强迫的部分去掉，变成好奇，把这个过程专注成思考和观察烟是什么？人为什么抽烟？
总之，习惯的养成和戒除，似乎都不是一蹴而就、有捷径可走的事情。为了让你在这个过程中稍微轻松一点，我们为你找了一些可以改变生活习惯的工具和方法论。
Read full article from 借助软件养成新习惯不容易，我们试了上百应用只有这 8 个坚持超过半年 | Hack Your Life – CodingBlog
Exclusive Time of Functions · OJ Games
Given the running logs of n functions that are executed in a nonpreemptive single threaded CPU, find the exclusive time of these functions.
Each function has a unique id, start from 0 to n-1. A function may be called recursively or by another function.
A log is a string has this format : function_id:start_or_end:timestamp. For example, "0:start:0" means function 0 starts from the very beginning of time 0. "0:end:0" means function 0 ends to the very end of time 0.
Exclusive time of a function is defined as the time spent within this function, the time spent by calling other functions should not be considered as this function's exclusive time. You should return the exclusive time of each function sorted by their function id.
Example 1:
Note:
- Input logs will be sorted by timestamp, NOT log id.
- Your output should be sorted by function id, which means the 0th element of your output corresponds to the exclusive time of function 0.
- Two functions won't start or end at the same time.
- Functions could be called recursively, and will always end.
- 1 <= n <= 100
This is Leetcode No.636. I think the most important thing to solve this problem is to understand the order of tasks running.
At first, I think I should count every subtask running time first, but I just need to calculate the first level subtasks.
So, I make the log tree first and calaculate them one by one.
Read full article from Exclusive Time of Functions · OJ Games