Question

我正在尝试使用C＃中的BayesServer library为Unity3D游戏构建动态贝叶斯网络。我有以下实现网络的方法

// numberOfDistractors and levelId will be used later for added complexity in modeling
void InitializeNetworkForLevel(int numberOfDistractors, int levelId)
{
    beliefNet = new BayesServer.Network();

    // add a knowledge node which is a latent variable (parameter to be learned from observed values
    KTrue = new State("KTrue");
    KFalse = new State("KFalse");
    knowledge = new Variable("Knowledge", KTrue, KFalse);
    knowledgeNode = new Node(knowledge)
    {
        TemporalType = TemporalType.Temporal // this is a time series node, hence re-used for each time slice
    };
    beliefNet.Nodes.Add(knowledgeNode);

    // add a question node, which denotes the oberved variable whether the question is answered correctly or not
    // this node has two states, namely correct or incorrect
    QTrue = new State("QTrue");
    QFalse = new State("QFalse");
    question = new Variable("Question", QTrue, QFalse);
    questionNode = new Node(question)
    {
        TemporalType = TemporalType.Temporal  // this is a time series node, hence re-used for each time slice
    };
    beliefNet.Nodes.Add(questionNode);

    // add a link from knowledge node to question node
    beliefNet.Links.Add(new Link(knowledgeNode, questionNode, 0));
    for (int i = 1; i <= 5; i++) 
        beliefNet.Links.Add(new Link(knowledgeNode, knowledgeNode, i)); // time series link (order/lag i)
    QueryNetwork(true);
}

然后是另一种推断方法：

void QueryNetwork(bool isAnswerCOrrect)
{

    StateContext kTrueTime0 = new StateContext(KTrue, 0);
    StateContext kFalseTime0 = new StateContext(KFalse, 0);

    Table priorKnowledge = knowledgeNode.NewDistribution(0).Table;
    priorKnowledge[kTrueTime0] = 0.5;
    priorKnowledge[kFalseTime0] = 0.5;
    // NewDistribution does not assign the new distribution, so it still must be assigned
    knowledgeNode.Distribution = priorKnowledge;

    // the second is specified for time >= 1
    Table learnRate = knowledgeNode.NewDistribution(1).Table;
    // when specifying temporal distributions, variables which belong to temporal nodes must have times associated
    // NOTE: Each time is specified relative to the current point in time which is defined as zero, 
    // therefore the time for variables at the previous time step is -1
    StateContext kTrueTime1 = new StateContext(KTrue, -1);
    StateContext kFalseTime1 = new StateContext(KFalse, -1);
    learnRate[kTrueTime1, kTrueTime0] = 0.5;
    learnRate[kFalseTime1, kTrueTime0] = 0.5;
    learnRate[kTrueTime1, kFalseTime0] = 0.5;
    learnRate[kFalseTime1, kFalseTime0] = 0.5;
    knowledgeNode.Distributions[1] = learnRate;

    Table answerStatus = questionNode.NewDistribution().Table;
    StateContext qTrue = new StateContext(QTrue, 0);
    StateContext qFalse = new StateContext(QFalse, 0);
    answerStatus[qTrue, kTrueTime0] = 0.5;
    answerStatus[qFalse, kTrueTime0] = 0.5;
    answerStatus[qTrue, kFalseTime0] = 0.5;
    answerStatus[qFalse, kFalseTime0] = 0.5;
    questionNode.Distribution = answerStatus;

    // optional check to validate network
    beliefNet.Validate(new ValidationOptions());
    // at this point the network has been fully specified

    // we will now perform some queries on the network
    RelevanceTreeInference inference = new RelevanceTreeInference(beliefNet);
    RelevanceTreeQueryOptions queryOptions = new RelevanceTreeQueryOptions();
    RelevanceTreeQueryOutput queryOutput = new RelevanceTreeQueryOutput();

    // set some temporal evidence
    if (isAnswerCOrrect)
        inference.Evidence.Set(question, new double?[] { 1, 0 }, 0, 0, 2);
    else
        inference.Evidence.Set(question, new double?[] { 0, 1 }, 0, 0, 2);

    queryOptions.LogLikelihood = true; // only ask for this if you really need it
    inference.Query(queryOptions, queryOutput); // note that this can raise an exception (see help for details)

    Debug.Log("LogLikelihood: " + queryOutput.LogLikelihood.Value);
}

但是，在尝试使用QueryNetwork方法验证网络时出现以下异常：

InvalidNetworkException：节点[知识]的分配为空。

BayesServer.Network.Validate（BayesServer.ValidationOptions选项）   （在：0处）

BayesNet.QueryNetwork（System.Boolean isAnswerCOrrect）（在   资产/脚本/BayesNet.cs:97）

BayesNet.InitializeNetworkForLevel（System.Int32 numberOfDistractors，   System.Int32 levelId）（在Assets / Scripts / BayesNet.cs：59处）

BayesNet.Start（）（位于Assets / Scripts / BayesNet.cs：21）

当我已经在QueryNetwork方法中指定知识节点时，为什么会说知识节点具有空分布。尽管我可以使用以下代码来解决此问题：

ValidationOptions opt = new ValidationOptions();
opt.AllowNullDistributions = true;
// optional check to validate network
beliefNet.Validate(opt);

此外，我假设第一级的所有概率为50％，如何根据第一级的推论更改这些值？

最终，我想构建如下图所示的网络，其中每个级别的Distractors数量都是不同的（如果过于复杂则可能是相同的）：

Answer 1

当我怀疑您在潜在节点上只需要滞后1时，您似乎在添加滞后1至5。尽管推理不是必需的，但为了进行测试，我建议在用户界面中展开网络以检查DBN是否符合您的期望。请注意，仅添加滞后1不会限制时间步长，只是每个步长仅连接到前一个时间步长。

BayesServer：InvalidNetworkException：节点[知识]的分配为空

1 个答案: