Prob Graph & Causal Graph

作者: shudaxu | 来源:发表于2021-01-25 17:47 被阅读0次

    Prob Graph: Probability Graph
    Causal Graph: Causal path diagrams
    差异:
    1、PGM中表示的是dependency,而非Causality。(这里很好理解,譬如Causal Inference Bias中Collider带来的伪关系。从统计学上确实是dependent,而确实也没有因果关系)
    2、PGM中,可以有DAG也可以有Undirected Graph。Causal Graph只能是DAG:bidirectional arrows and feedback loops are not permitted
    3、目的不同。当我们用PGM来做预测(probability inference,prediction)的时候,因为我们只关系预估是否准确[2],大多数情况只需要将所有的factor都纳入即可。而做Causal Inference的时候,基本是考虑单变量影响,其他的变量是用来prevent confounding of the causal path。

    Refer:
    [1]:From Dependency to Causality: A Machine Learning Approach
    [2]:When is a confounder not a confounder?

    相关文章

      网友评论

        本文标题:Prob Graph & Causal Graph

        本文链接:https://www.haomeiwen.com/subject/rkiqzktx.html