摘要:时间序列分析之holtwinters和ARIMA https://blog.csdn.net/huangtiao2509/article/details/78251101 时序算法,常谈平稳性,弱平稳性,强平稳性 平稳的基本思想是:时间序列的行为并不随时间改变。平稳性刻画的是时间序列的统计性质关于时
阅读全文
摘要:https://blog.csdn.net/u010665216/article/details/78051192 mark
阅读全文
摘要:Understanding Variational Autoencoders (VAEs) 为何不能用AE的decoder来直接生成数据? 因为这里的latent space的regularity无法保证 右边给出的例子,AE只是保证training过程中的cases的这些离散点,会导致严重的ove
阅读全文
摘要:https://zhuanlan.zhihu.com/p/28522637 https://zhuanlan.zhihu.com/p/21930884 mark
阅读全文
摘要:看下用户注册StateMachine的过程, MapStateMachine::new这会构造一个supplier 在build中,传入初始化ServerContext ServerContext中, reset逻辑中, 这里看到stateContext的定义, 也是一个单线程,所以这里有两个thr
阅读全文
摘要:Copycat server之间的configure是如何,何时被同步的? 大家可以看到,只有leader可以同步配置 1. 显式的调用LeaderState.configure LeaderState.configure /** * Commits the given configuration. */ protected CompletableFutu...
阅读全文
摘要:对于Command,Configuration都要通过appendEntries的方式,把Entries同步给follower LeaderState.configure /** * Commits the given configuration. */ protected CompletableFuture configure(Collection members) { ...
阅读全文
摘要:Member.Status status的变迁是源于heartbeat heartbeat,append空的entries heartbeat的逻辑是会向所有的getRemoteMemberStates,发起heartbeat AVAILABLE 在初始化的时候,每个ServerMember默认是S
阅读全文
摘要:client.submit(new PutCommand("foo", "Hello world!")); ServerContextconnection.handler(CommandRequest.class, request -> state.command(request)); State.command ReserveState开始,会把command forward到leade...
阅读全文
摘要:Server被拉起有两种方式, Address address = new Address("123.456.789.0", 5000); CopycatServer.Builder builder = CopycatServer.builder(address); builder.withStateMachine(MapStateMachine::new); 自己拉起一个cluster, ...
阅读全文
摘要:Copycat’s primary role is as a framework for building highly consistent, fault-tolerant replicated state machines. Copycat servers receive state machi
阅读全文
摘要:https://github.com/atomix/copycat http://atomix.io/copycat/docs/membership/ 为了便于实现,Copycat把member分成3种, active, passive, and reserve members — each of which play some role in supporting rapid ...
阅读全文
摘要:http://commons.apache.org/proper/commons-math/userguide/stat.html mark DescriptiveStatistics maintains the input data in memory and has the capability of producing "rolling" statistics computed f...
阅读全文
摘要:http://www.deeplearningbook.org/ mark
阅读全文
摘要:Amazon’s Dynamo [9] and Facebook’s Cassandra [13], relax the consistency model,and offer only eventual consistency. Others such as HBase [1] and BigTable [4] offer strong consistency only for operat...
阅读全文
摘要:参考, http://data-artisans.com/high-throughput-low-latency-and-exactly-once-stream-processing-with-apache-flink/ http://www.confluent.io/blog/real-time-stream-processing-the-next-step-for-apache-flink...
阅读全文
摘要:HyperLogLog参考下面这篇blog, http://blog.codinglabs.org/articles/algorithms-for-cardinality-estimation-part-iv.html 为何LLC在基数不大的时候会误差比较大? 直观上,由于基数不大时,会有很多空桶,而最终结果是求平均值,这个值对离群值(这里的0)非常敏感 那么重理论上看,为何误差比较...
阅读全文
摘要:http://www.cs.berkeley.edu/~matei/papers/2013/sosp_sparrow.pdf http://www.eecs.berkeley.edu/~keo/talks/sparrow-sosp-talk.pdf 解决的问题 现有的scheduler方案, 都是基于master的, 因为schedule必须要知道所有slave的情况, 然后才能决定...
阅读全文
摘要:http://highlyscalable.wordpress.com/2013/08/20/in-stream-big-data-processing/ Overview In recent years, this idea got a lot of traction and a whole bunch of solutions like Twitter’s Storm, Yahoo’...
阅读全文
摘要:Probabilistic Data Structures for Web Analytics and Data Mining 对于big data经常需要做如下的查询和统计, Cardinality Estimation (基数或势), 集合中不同元素的个数, 比如, 独立访客(Unique Vi
阅读全文