QxRed

虽趣舍万殊，静躁不同，当其欣于所遇，暂得于己，快然自足，不知老之将至；及其所之既倦，情随事迁，感慨系之矣。向之所欣，俯仰之间，已为陈迹，犹不能不以之兴怀，况修短随化，终期于尽！古人云，“死生亦大矣。”岂不痛哉！

人的一生，用各种短暂的满足来忘却生死，悲啊！

http://www.fjdh.com/main/musonline/mus/diandeng.mp3

点一盏心灯 照亮黑暗的心灵角落
点一盏心灯 带来希望的每一分钟燃起的火焰
点一盏心灯 照亮黑暗的心灵角落
点一盏心灯 带来希望的每一分钟燃起的火焰
一朵 两朵 三朵 千朵 万朵 留给哀伤的泪眼
一朵 两朵 三朵 千朵 万朵 留给迷路的旅人
点一盏心灯 照亮黑暗的心灵角落
点一盏心灯 带来希望的每一分钟燃起的火焰
一朵 两朵 三朵 千朵 万朵 留给哀伤的泪眼
一朵 两朵 三朵 千朵 万朵 留给迷路的旅人
一朵 两朵 三朵 千朵 万朵 献给哭泣的弱者
一朵 两朵 三朵 千朵 万朵 献给苦痛的众生
点一盏心灯 带来希望的每一分钟……

方法很新颖，结果比别人差。
这种文章真臭美。
没有精度，就没有发言权，就不要出来献丑。
还分析了半天原因，咯里啰唆。
精度上不去，任何合理的方法都是不合理的，
只要精度上去，不合理的方法都是合理的。

本科毕业后，开始接触机器学习。这是一个有趣的领域，让整天只会打游戏的我对学习产生兴趣。这些年确实学到了很多东西，让我了解了一些很有前景的技术。但是，我并不打算继续下去，以论文为中心的学术界已经脱离了实际应用，作为一个学生，我更希望所学可以改变人们的日常生活，就象Ajax那样。我希望前景可以变为现实。今后我会偏向于工程方面，尽我的努力把机器学习这个技术的魅力发挥出来。

由于没有javascript脚本调试工具，只能靠输出来看见变量内容。以下是我想看函数的形参值的代码：
<script language="javascript">
function chk(i)
{
    location.href=i;
}
。。。

sourceforge 镜像：
https://sourceforge.net/project/showfiles.php?group_id=201943&package_id=240376&release_id=672852

google code镜像：
http://code.google.com/p/pocketcrf/
增加了passive aggressive训练算法
支持范围的特征，比如一个词的周围的词集的特征

EQ测试：(满分20分)
http://astro.sina.com.cn/t/2004-12-27/15038.shtml

您的成绩：10分
你意识到自己和他人的情感，但有时却忽视它们，你不重视它们，你喜欢比较现实的东西，达到感观上的满足你就觉得够了，你忽视心灵 层次的需要，不太会做情感方面的沟通和交融，但你却不明白这对你的幸福而言是多么重要。事实上你对乔迁和你的股票大涨等诸如此类的事情更有兴趣，你比较在 乎的是物质方面的收益，这些几乎支配着你的生活，也是你生活的重心。然而，无论实现多少物质目标，你仍然感到不满足，如果你的生活只是围绕着它的，那么你 永远也不能让自己停歇下来。试着去分析和理解你的情感，感觉你内心最真实的一角，并且按照你真正想要的去行动，你会更幸福。记住，有时可能会被暂时压制， 但是，你总是能够从挫折中吸取教训而不是消沉，并能重新创造你的优势，让自己变得更完美。

EQ低是很糟糕的事情，在交流的时候会出现很大的问题。
EQ高低是天生的么，就像IQ一样？

chapter 2
1.Conditional independence, is X_A⊥X_B | X_C ?
  (1) directed graphs: bayes ball algorithm, page 17
          shade X_C, place a ball at each of the nodes X_A, let the balls bounce around according to 3 rules (figure 2.9-figure 2.13). If none of the balls reach X_B, then X_A⊥X_B | X_C; otherwise independence is not true
  (2) undirected graphs: X_A⊥X_B | X_C holds if and only if all paths from X_A to X_B pass through X_C (figure 2.21)

chapter 3
1.Elimination algorithm
  (1) directed graphs: figure 3.3
  (2) undirected graphs: the only change occures in the INITIALIZATION stage, replace local conditional probabilities by potentials, page 11
 
2.Complexity analysis
  (1) Graph elimination algorithm
      (a) undirected graphs: connect all the remaining neighbors of X_i, remove X_i from the graph (figure 3.5, 3.6)
      (b) directed graphs: convert to undirected graph by moralization (figure 3.9, 3.10), then use the same algorithm for undirected graphs
  (2) reconsistituted graph: retain the edges created during the elimination procedure (figure 3.7)
  (3) complexity:
      (a) exponential in treewidth: one less than the smallest achievable value of the cardinality of the largest elimination clique over all possible elimination orders, page 16
      (b) finding the best elimination order is NP hard


chapter 4
1. Message passing
   (1) Message passing (equation 4.7, 4.8)
   (2) Message passing protocol page 9
2. Sum-Product algorithm
   (1) synchronous parallel algorithm, page 9, paragraph next to Message passing protocol (figure 4.4)
   (2) Inward Outward algorithm for tree, figure 4.5
3. Factor graph
   (1) directed trees, treat each conditional probability as a pontential, construct factor graph as undirected graphs
   (2) Message passing (equation 4.12, 4,13, 4.16-4.19)
   (3) Message passing protocol, page 18
   (4) Sum Product algorithm for tree, figure 4.11
   (5) Polytrees: replace each v structure by a factor, page 24 (figure 4.15)
4. Max product algorithm, replace sum operation by max operation

chapter 16
1. properties of conditional indendence. (theorem 4.1-4.5)
2. relations between properties of distribution for directed graphs:
   DF <=> DL <=> DG <=> d-separation => DP; DP => d-separation if all probabilites are positive
   (1) DF: directed factorization. P(S)=\prod_i P(S_i|pa(S_i)). (equation 4.16)
   (2) DL: directed local markov. any node is conditionally independent of its non-descendents given its parents: a⊥nd(a)|pa(a), page 77 (equation 4.22)
   (3) DG: directed global markov. A⊥B|C whenever C separates A from B in the smallest anncestral moral graph (smallest anncestral moral graph, figure 4.17)
   (4) d-separation: P(S) satisfies all conditional independence implied by the graph  
   (5) DP: directed pairwise markov: for any two nodes a,b , b∈nd(a), we have a ⊥ b | nd(a)\b (figure 4.18, equation 4.24)
3. relations between properties of distribution for undirected graphs:
   F => G => P <=> L; F <=> G <=> P <=> L if all probabilites are positive
   (1) F: P(S)=\proc_c \phi(c)
   (2) G: global markov. For any disjoint subsets of nodes A,B,C such that C separates A,B, P(S) satisfies A⊥B|C
   (3) P: pairwise markov. For any two nodes a,b in the graph such that there is no direct link in the graph from a to b, then a⊥b | S\{a,b}
   (4) L: local markov. a⊥S\cl(a) | bd(a)
4. relations between directed graph and its moral graph
   (1) independence in moral graph => independence in directed graph (theorem 4.8)
   (2) independence in directed graph => independence in smallest ancestral moral graph (theorem 4.9)
  
chapter 17
1.Junction tree
   (1) Property: for any pair of cliques V and W, all cliques on the unique path between W and V contain W∩V , page 22 (figure 17.3)
   (2) local consistency
       (a) \phi(S)=\sum_{W\S} \phi(W), for any W ∈ N(S)
       (b) two pass algorithm (equation 17.14, 17.15, 17.18, 17.19)
       (c) Propagation in a clique tree, Message passing protocol, page 18 (figure 17.11)
   (3) global consistency (equation 17.11)
   (4) construction algorithm: Maxmimal spanning tree algorithm (figure 17.14)
   (5) Hugin algorithm for deriving equation 17.11, page 29
2 Triangulation
   (1) A graph is triangulated if there are no chordless cycles in the graph, page 25
   (2) All triangulated graph have a junction tree (theorem 3)
   (3) graph elimination for undirected graph yield a trangulated graph. (theorem 4)
   (4) Decomposable = triangulated = junction tree ,page 36 (theorem 7)

2，3，4很正统，从第五章开始就是分类器之类的东西了，EM也在里面。额，好像连重要的Loopy都没有
Jordan 啊 Jordan，偶想告诉你，如果要我再选一次，我就不看你的书了

上海上空一整块臭氧层缺失，强烈的紫外线直射地面，造成数百度的气温。暖空气迅速上升，强烈的气压差使得冷空气从四面八方补充过来，形成漩涡。
复旦双子楼，我伫立在西塔的高处，透过滚烫的玻璃向外张望，突然间整块玻璃因室内外的巨大温差而轰然碎落，强大的旋风把我抛到了一块坚硬的墙壁上。我跌跌撞撞地站起来，到处是奔走呼救的声音，我随着人群涌向塔的内侧，那里有安全出口。
由于楼梯太窄，我们在楼梯口焦急地等着、推着。
就在这时，强大的旋风把西塔从中间横腰切断，塔顶向内侧倾斜。我感觉到后面的人渐渐压在了我背上，突然间整个楼层轰然滑落，开始自由落体。我们被地压在玻璃窗上往下坠落。透过玻璃，我看见东塔的人们扶着窗户，用惊恐的眼神望着我们。
塔塌了一半，突然停住，我们随着玻璃继续往下，被几个支出的墙壁碰撞缓冲之后，我被重重地砸在了地上。我意识到，自己还可以动。于是爬起来。这里是双子楼底楼的中央大厅，到处都是灰尘废墟，无数的玻璃窗被大风吹得在大厅中央横向穿梭。
我打算从楼北面出口逃走，可是那里已经被巨大的断壁封锁，于是转向南面正门逃生。刚出大门，发现外面是一片耀眼的白光刺得人几乎睁不开眼，除了倾斜的大楼的轮廓外，什么都看不见。高温烫的我触电式地缩回来，这时一块宽5米多的巨大玻璃横着像我迎面飞来。

我猛然惊醒，南极洲！南极洲！