概率图模型
Baocai3000
技术源于需求又高于需求
展开
-
概率图模型【读书笔记1】概率理论
对于概率probability的两种理解:(1)频率,frequencies of events,无法解决比如“明天下雨概率是0.8”,因为无法定义频率,明天只有一个。(2)信念的接受程度,subjective degree of belief.原创 2011-01-23 23:36:00 · 1398 阅读 · 0 评论 -
Conditonal Bayesian Network
翻译 2011-09-26 16:07:00 · 608 阅读 · 0 评论 -
Naive Bayes Model
纯粹来自于对于因果关系的直觉翻译 2011-09-26 15:06:55 · 1131 阅读 · 0 评论 -
对Bayes网络的查询
查询内容;1.预测边缘概率(Marginal Inference)2.MAP 推测(argmax(f,a)P(F=f,A=a|H=t))3.活动数据收集(如何确定下个查询是什么,在上两个任务之上)查询以及推理的pattern:1.结果推理或者预测:do翻译 2011-09-26 15:31:41 · 486 阅读 · 0 评论 -
Reality vs. Model
翻译 2011-09-26 15:36:18 · 548 阅读 · 0 评论 -
Bayes网络独立性的来源
1.从真实的概率P得到(不可能实现);2.为了方便;3.因果关系的直觉4.认真的搜索(结构学习)翻译 2011-09-26 15:04:25 · 622 阅读 · 0 评论 -
D分割
翻译 2011-09-26 15:42:17 · 1354 阅读 · 0 评论 -
Tree CPD
翻译 2011-09-26 15:57:55 · 609 阅读 · 0 评论 -
covariance
In probability theory and statistics, covariance is a measure of how much two variables change together. Variance is a special case of t翻译 2011-09-29 11:24:38 · 701 阅读 · 0 评论 -
sum rule
翻译 2011-09-29 10:21:00 · 980 阅读 · 0 评论 -
贝耶斯
贝耶斯网络的独立性假设:1.局部马尔科夫假设:变量X,只给定x的父亲,则x独立于其非子孙节点。翻译 2011-09-26 14:59:07 · 663 阅读 · 0 评论 -
Discriminative models, not discriminative training
<br /><br />原创 2011-04-21 23:17:00 · 1072 阅读 · 0 评论 -
token转图
<br />输入,一个instance的信息,其中最后一个为target:<br /> <br />重庆 ns NON_PREFIX NON_SUBJECT地区 n NON_PREFIX NON_SUBJECT <br />假设使用的conjection为:<br /> int[][] conjunctions = new int[3][]; conjunctions[0] = new int[] { -1 }; conjunctions[1] = new int[] { 1 };原创 2011-04-03 01:19:00 · 926 阅读 · 1 评论 -
context-specific representation
<br />两种表示方法:tree和rule表示方法<br />============================<br />今天元宵节,买了10块钱的元宵,20个,然后去食堂代煮了一下,又花了3块钱。然后去看烟花,拍了几张照片,回来看元宵晚会,很爽啊。看书看到现在,困困的。<br />原创 2011-02-18 01:09:00 · 752 阅读 · 0 评论 -
存储以及分析图结构
<br />1.This was changed by Google in October 2010 through the presentation of their live search index system Caffeine[42] which is based on the scalable infrastructure Pregel. Pregel allows to dynamically add and update nodes in the system and the added o原创 2011-03-12 10:06:00 · 746 阅读 · 0 评论 -
概率图模型【读书笔记2】重读
<br />1.1Motivation:<br />1.Declarative representation, model-based methods, which separate data from algorithms which apply to them.<br />2.above is a fundamental component in many fields,especially in uncertainy.Uncertainty is prevalence in the real worl原创 2011-03-28 10:07:00 · 983 阅读 · 0 评论 -
利用反射机制调用main函数并传递参数
<br />public void testGenMallet() { try { Method methods = Csv2Vectors.class .getMethod("main", String[].class); methods.invoke(null, (Object) new String[] { "--input", "data/train_10000.data", "--output", "web.mallet", "--pres原创 2011-03-29 00:01:00 · 1171 阅读 · 0 评论 -
MALLET中的序列化
<br />public void write(File f) { try { ObjectOutputStream oos = new ObjectOutputStream( new FileOutputStream(f)); oos.writeObject(this); oos.close(); } catch (IOException e) { System.err.println("Exception writing file " + f +原创 2011-03-31 00:17:00 · 1166 阅读 · 0 评论 -
获取LABEL与构建InstanceList
<br />String[] labels = new String[crf.getOutputAlphabet().size()]; for (int i = 0; i < crf.getOutputAlphabet().size(); i++) { labels[i] = (String) crf.getOutputAlphabet().lookupObject(i); } <br /> public void testCRF() throws Exception { Pipe原创 2011-03-31 09:00:00 · 1056 阅读 · 0 评论 -
MALLET预测
<br />public void testPredict() throws Exception { CRF crf = (CRF) FileUtils.readObject(new File("web.model")); Pipe p = makeSpacePredictionPipe(); InstanceList predictInstances = new InstanceList(p); predictInstances.addThruPipe(new LineGroupI原创 2011-03-31 10:33:00 · 949 阅读 · 0 评论 -
MALLET中instance的data
<br /> <br />1.在public Instance pipe(Instance carrier)的构造函数中,打印部分注释为:<br />Swapping order, since data often has a newline at the end -DM <br />2.滑动窗口<br />newFeature += (i == 0 ? "" : "_&_") + s + (conjunctions[j][i] == 0 ? "" : ("@"原创 2011-04-02 20:29:00 · 1074 阅读 · 0 评论 -
生成FeatureVector
/** * Recursively makes conjunctions by iterating through features at each * offset * * @param iters * iterate over the PropertyLists at each offset * @param currIndex * which offset we're currently on, e..g 1 i原创 2011-04-03 00:13:00 · 1228 阅读 · 0 评论 -
复杂网络
//http://zh.wikipedia.org/wiki/%E5%A4%8D%E6%9D%82%E7%BD%91%E7%BB%9C1.复杂网络,复杂网络是由数量巨大的节点和节点之间错综复杂的关系共同构成的网络结构。2.小世界网络:小世界网络的判定准则有两个,分别是特征路径长度短,和高集聚系数。网络的特征路径长度是指在它的图表示中,两个节点的路径长度的平均值(这里路径长度指两节点间最短路转载 2012-11-13 11:27:09 · 2421 阅读 · 0 评论