# 统计代写|贝叶斯分析代写Bayesian Analysis代考|MAST90125

## 统计代写|贝叶斯分析代写Bayesian Analysis代考|Multiobject Bayesian Network Models

There are two types of situations in which it becomes inefficient or impractical to model a problem using a single $\mathrm{BN}$ :

1. When the model contains so many nodes that it becomes conceptually too difficult to understand. One of the key benefits of $\mathrm{BNs}$ is that they are such a powerful visual aid. But, unless there is some especially simple structure, as soon as a model contains more than, say, 30 nodes its visual representation becomes hard to follow. It needs to be broken up into smaller understandable chunks.
2. When the model contains many similar repeated fragments, such as the case when there are repeated instances of variable names that differ only because they represent a different point in time. For example, the model in Figure $8.48$ (which represents a sequence of days in which we assess the risk of flood) contains several such repeated variables that differ only by the day they are recorded.

In both cases we seek to decompose the model into smaller component models. For example, the large model shown in Figure $8.49$ ought to be decomposable somehow into three smaller models as indicated by the grouping of the nodes, whereas the flood model ought to be decomposable into a sequence of identical models of the type shown in Figure 8.50.

The component models we need are called object-oriented BNs (OOBNs), because they have some of the properties associated with object-oriented modeling.

An OOBN is simply a BN with certain additional features that make it reusable as part of a larger $\mathrm{BN}$ model. The most important feature of an OOBN is that it will generally have input and/or output nodes. These represent the “external interface” of the BN and enable us to link OOBNs in a well-defined way.

For example, the model shown in Figure $8.49$ can be decomposed into three OOBNs with input and output nodes as shown in Figure 8.51.
In a tool like AgenaRisk the individual OOBNs can be embedded and linked into a higher-level model as shown in Figure 8.52. You need to specify which nodes in an OOBN are input and output nodes; at the higher level only the input and output nodes are shown (so the OOBN “Test Quality” has a single output node called test quality and the OOBN “Reliability During Test” has a single output node called reliability during test). We can then simply link relevant pairs of input and output nodes in two different OOBNs as shown in Figure 8.52.

## 统计代写|贝叶斯分析代写Bayesian Analysis代考|The Missing Variable Fallacy

One of the advantages of using Bayesian networks is that it encourages domain experts to articulate and make visible all their assumptions and then ensures they are carefully modeled. One way to determine whether your model has really articulated all of the necessary assumptions is to build and run the model and determine whether the conclusions make sense.

The missing variable fallacy is explained by the following example. Suppose it is known that, on average, $50 \%$ of the students who start a course pass it. Is it correct to conclude the following?
a. A course that starts with 100 students will end up, on average, with 50 passes.
b. A course that ends with 50 passes will, on average, have started with 100 students.
In fact, although the first conclusion is normally correct you may be very surprised to learn that the second conclusion is normally not true. This has everything to do with the way we reason with prior assumptions, which, as we have seen, lies at the heart of the Bayesian approach to probability.

To explain how this fallacy results in the wrong BN model (and how to fix it) we will add some more detail to the example. The crucial prior assumption in this case is the probability distribution of student numbers who start courses. Let’s suppose that these are courses in a particular college where the average number of students per course is 180 . We know that some courses will have more than 180 and some less. Let’s suppose the distribution of student numbers looks like that in Figure 8.57. This is a Normal distribution whose mean is 180. As we have already seen in Chapter 2, Normal distributions are characterized not just by the mean but also by the standard deviation, which is how spread out the distribution is. In this example the standard deviation is 20 .
Because the number of students who pass is obviously influenced by the number who start, we represent this relationship by the $\mathrm{BN}$ shown in Figure 8.58(a).

## 统计代写|贝叶斯分析代写Bayesian Analysis代考|Multiobject Bayesian Network Models

1. 当模型包含如此多的节点时，它在概念上变得难以理解。的主要好处之一乙ñs是它们是如此强大的视觉辅助工具。但是，除非有一些特别简单的结构，否则一旦模型包含超过 30 个节点，它的视觉表示就会变得难以理解。它需要分解成更小的可理解的块。
2. 当模型包含许多相似的重复片段时，例如变量名称的重复实例仅因为它们代表不同的时间点而不同。例如图中的模型8.48（代表我们评估洪水风险的一系列日子）包含几个这样的重复变量，它们仅在记录的日期不同。

OOBN 只是一个具有某些附加功能的 BN，使其可作为更大的一部分重复使用乙ñ模型。OOBN 最重要的特征是它通常具有输入和/或输出节点。这些代表 BN 的“外部接口”，使我们能够以明确定义的方式链接 OOBN。

## 统计代写|贝叶斯分析代写Bayesian Analysis代考|The Missing Variable Fallacy

myassignments-help数学代考价格说明

1、客户需提供物理代考的网址，相关账户，以及课程名称，Textbook等相关资料~客服会根据作业数量和持续时间给您定价~使收费透明，让您清楚的知道您的钱花在什么地方。

2、数学代写一般每篇报价约为600—1000rmb，费用根据持续时间、周作业量、成绩要求有所浮动(持续时间越长约便宜、周作业量越多约贵、成绩要求越高越贵)，报价后价格觉得合适，可以先付一周的款，我们帮你试做，满意后再继续，遇到Fail全额退款。

3、myassignments-help公司所有MATH作业代写服务支持付半款，全款，周付款，周付款一方面方便大家查阅自己的分数，一方面也方便大家资金周转，注意:每周固定周一时先预付下周的定金，不付定金不予继续做。物理代写一次性付清打9.5折。

Math作业代写、数学代写常见问题

myassignments-help擅长领域包含但不是全部: