explianit-cisco-2019-sigmod

causation理論的一點應用。證實分值不是偶然發生的

RCA的工具通常能夠query and classify anomalies，相關性分析（causal probabilistic gaphical models）

• spurious correlations。當dimensionality比data points數量多
• 交互式查詢，target metrics of interest(Y),正常和異常時間段，specificity metrics for control(可選Z)，search space of metrics(可選X)=》TOP 20 root-cause insearchspace：scores(Xi)<-assoc(Y,Xi|Z)

原理

causal bayesian network。嗯，能夠用帶條件的兩個變量關係去構造複雜的關係。

- ExplainIt!– A Declarative Root-cause Analysis Engine for Time Series Data
 - Why? The above approach offers three main benefits. 
 - First, the formalism is a non-parametric and declarative way of expressing dependencies between variables and defers any specific approach to the runtime system. 
 - Second, the unified approach naturally lends itself to multivariate dependencies of more complex relationships beyond simple correlations between pairwise univariate metrics. 
 - Third, the approach also gives us a way to reason about dependencies that might be easier to detect only when holding some variables con- stant;

1.feature family （能夠按照host聚合，相似group by。好比某個feature family是75th延時，當前clusterjobs數量）

2.ranking 假設（X,Y,Z）=》給出Xi的排序
單變量Z空score:X中每一個Xi,Y中每一個Yj,Pearson product-moment coorelation 的均值和最值 coorMean=meani,j|pi,j|。
多變量Z空，線性迴歸（random projection降維）+loss function 計算R方
Z不空：迴歸Y~Z,X~Z.獲得RY;X.,RX;Z. 迴歸兩個R計算R2(Y;X|Z)
當X中predictors不少，observations不多時。用Ridge penalty達到了和adjusted R2同樣的效果。見後文。

實驗是否可以補全圖

評估

打分方法的評估：
ranking accuracy：cause是第r個，1/r
success rate: cause in topk 得1，不然0

理論

PC/SGS算法 use pairwise conditional independence=>full causal structture.also considering a joint set of variables.
rarely requires the full causal structuew

給出了過擬合 用radj。當一個score至少大於s是意外正常發生的機率和n,p的關係。當s小於這個值時不可信的。