z*l
发帖数: 30 | 1 if it overestimates and can lead to a suboptimal solution then it is NOT a
good heuristic.
Try to think about some very naive heuristic, e.g. just like bfs
to |
|
g*****n
发帖数: 239 | 2 Any good heuristic for 8 puzzle that sometimes overestimates and can lead to
a suboptimal solution? |
|
H*7
发帖数: 4894 | 3 ☆─────────────────────────────────────☆
mapcar (再次等待 - 明年) 于 (Mon Mar 2 18:18:36 2009) 提到:
下棋的时候, 大家都在比算路. 你可以比对手多算一步, 你就牛比不少. 象我, 下象
棋的时候, 一般都算一步. 然后在我的心中有一个heuristic, 然后就用heuristic来
比较, 然后就下了. 然后, 牛比的人当然就把我砍了. 但是我可以砍没有heuristic
的人, 就算他比我算的多.
打球的时候, 是不是也有这个问题呢? 比如, 双打的基本战术就是一个heuristic.
然后比较细节的地方呢, 牛比的人就有预谋. 一个典型的例子就是, 网前的人可以放
一个高质量的网前球, 然后他计算中的下一步, 就是对方跳起来, 然后他搭档杀死.
☆─────────────────────────────────────☆
Riemann (幸福的曼何) 于 (Mon Mar 2 18:26:49 2009) 提到:
第一步 啊球打过去
第二步放网, 期待对方... 阅读全帖 |
|
z****t
发帖数: 4322 | 4 昨天在飞机上碰到了以前的强森教授，老E给我从deterministic optimization的角度
讲解了一下目前的美国经济形势。
我老灵感大发，决定用operation research的角度讲解一下中国足球，给中国足球解解
盘。这个帖子可能只有学数学和统计优化的能看懂，如果讲解不对zaoqi,drake等内行
也不要嘲笑。
如果说中国足球是一mix integer program的话,我们现在最需要的是an exact
solution to the MIP. 我们山东人前些年就是在用pure enumeration的方式来解，但
是南方人特别是江浙沪一带的却一直在用heuristic来解问题。南方的问题是，从来不
会100%的努力，而总是想些捷径。在场上就知道用脑子来想最优的办法，而不是努力，
努力再努力。就像我说的，南方人上场踢球都带着CPLEX。 就像很多用cplex的人一样
，总是忽略了cplex虽然牛，但是却有很多limitation, 有warm up找到first feasible
solution的时间，可能比我们北方人搞pure enumeration还要时间... 阅读全帖 |
|
m***n
发帖数: 12188 | 5 你这个跳跃太大了。
计算机出现前，无法理性处理和分析大数据，但是可以通过 heuristic 来处理。
其实人脑和动物脑的circuits就是在演化中建立了许多的 heuristic circuits，才使
得生存成为可能。
比如人在森林里面看见无风草动就自然毛骨悚然，无须思考，直接进入战斗或者逃跑模
式。在大脑反应过来之前，肾上腺素已经开始疯狂分泌，肠胃停止蠕动，皮肤表层毛细
血管收缩（受伤后可以减少流血和能量损失），痛觉神经中枢沉默（可以战斗得更有效
率），心跳加速，肝脏疯狂活跃起来，无关思维活动全部压抑下去，如果这个时候作一
个大脑MRI, 就会发现对未来规划的区域活动完全停止，从爬行类时代就进化出的
circuits 疯狂活跃起来，接管一切。这是你无数代祖先和大猫战斗的经验，包括目睹
同伴被大猫吃掉的经验所产生的 heuristic。
另外一个例子是视觉，用电脑分析图片，辨识物体是很漫长的计算，效果还不很好，但
是人脑早就形成了无数 heuristic，直接看出远近，大小，固体，流体，物体，几乎不
需要时间。
易经之类的占卦也是一种 heuristics。按照我的个人经验，如果... 阅读全帖 |
|
w****i
发帖数: 964 | 6 MCT is basically a sampling method, MCT with random policy performs very
poorly. The existing MCT approaches before AlphaGO heavily rely on hand
crafted heuristics in their policies. This is exactly the reason why MCT
approaches basically stop going forward after reaching KGS 5-6d, for almost
5 years, because hand crafted heuristic hits the limit. The key component of
the new program is the policy/value network to automatically construct
heuristics through machine learning. The major player here... 阅读全帖 |
|
m****r
发帖数: 6639 | 7 下棋的时候, 大家都在比算路. 你可以比对手多算一步, 你就牛比不少. 象我, 下象
棋的时候, 一般都算一步. 然后在我的心中有一个heuristic, 然后就用heuristic来
比较, 然后就下了. 然后, 牛比的人当然就把我砍了. 但是我可以砍没有heuristic
的人, 就算他比我算的多.
打球的时候, 是不是也有这个问题呢? 比如, 双打的基本战术就是一个heuristic.
然后比较细节的地方呢, 牛比的人就有预谋. 一个典型的例子就是, 网前的人可以放
一个高质量的网前球, 然后他计算中的下一步, 就是对方跳起来, 然后他搭档杀死. |
|
w****i
发帖数: 964 | 8
Apparently you have no idea about the difference. previous heuristics are
hand-crafted heuristics. It's the DCNN learned heuristics that offers a well
balanced policy in searching. The difference is how to construct the
heuristics, by hand or by machine.
So obviously you do not know the paper and are blind to the fact that a pure
DCNN bot have reached ~KGS 3d level, when you are arguing with bunch of
people with your limited (and outdated) knowledge of computer Go, you are
the one looks really ... 阅读全帖 |
|
d********e
发帖数: 132 | 9 论文中每章 都包括一些 algorithm，heuristic 或者procedure。我用algorithms
package 写这些algorithm，heuristic 或者procedure。 在caption栏我想要的编号
方式是：
Chapter 1
Algorithm 1.1， Algorithm 1.2 Procedure 1.1 Procedure 1.2
Chapter 2
Algorithm 2.1 Algorithm 2.2 Heuristic 2.1 Procedure 2.1 Procedure 2.2
但现在按照default 的方法生成的编号是完全是不分章节和名称的统一编号：
Algorithm 1， Algorithm 2 , Procedure 3, Procedure 4, Algorithm 5, Algorithm
6, Heuristic 7, Procedure 8, Procedure 9
请问一下怎么实现我想要的编号方式。
谢谢帮忙！ |
|
d********e
发帖数: 132 | 10 论文中每章 都包括一些 algorithm，heuristic 或者procedure。我用algorithms
package 写这些algorithm，heuristic 或者procedure。 在caption栏我想要的编号
方式是：
Chapter 1
Algorithm 1.1， Algorithm 1.2 Procedure 1.1 Procedure 1.2
Chapter 2
Algorithm 2.1 Algorithm 2.2 Heuristic 2.1 Procedure 2.1 Procedure 2.2
但现在按照default 的方法生成的编号是完全是不分章节和名称的统一编号：
Algorithm 1， Algorithm 2 , Procedure 3, Procedure 4, Algorithm 5, Algorithm
6, Heuristic 7, Procedure 8, Procedure 9
请问一下怎么实现我想要的 编号方式。
谢谢帮忙！ |
|
d********e
发帖数: 132 | 11 论文中每章 都包括一些 algorithm，heuristic 或者procedure。我用algorithms
package 写这些algorithm，heuristic 或者procedure。 在caption栏我想要的编号
方式是：
Chapter 1
Algorithm 1.1， Algorithm 1.2 Procedure 1.1 Procedure 1.2
Chapter 2
Algorithm 2.1 Algorithm 2.2 Heuristic 2.1 Procedure 2.1 Procedure 2.2
但现在按照default 的方法生成的编号是完全是不分章节和名称的统一编号：
Algorithm 1， Algorithm 2 , Procedure 3, Procedure 4, Algorithm 5, Algorithm
6, Heuristic 7, Procedure 8, Procedure 9
请问一下怎么实现我想要的 编号方式。
谢谢帮忙！ |
|
o**********e
发帖数: 18403 | 12
re. 孝was a heuristic, widely used during
Han dynasty, in order to educate and select
trustworthy bureaucrats (忠).
Examination is the 2nd heuristics, in order to
supplement and improve 忠孝's results.
敝帚自珍明珠蒙尘(VIII) 孝文化绝对珍贵
http://www.mitbbs.com/article_t/History/32145081.html
The Chinese heuristics were unique and
spectacularly successful among pre-industrial
societies. I don't understand why post-industrial Chinese
or anybody could look down their noses on the ancient Chinese. |
|
o**********e
发帖数: 18403 | 13
re. 孝was a heuristic, widely used during
Han dynasty, in order to educate and select
trustworthy bureaucrats (忠).
Examination is the 2nd heuristics, in order to
supplement and improve 忠孝's results.
敝帚自珍明珠蒙尘(VIII) 孝文化绝对珍贵
http://www.mitbbs.com/article_t/History/32145081.html
The Chinese heuristics were unique and
spectacularly successful among pre-industrial
societies. I don't understand why post-industrial Chinese
or anybody could look down their noses on the ancient Chinese. |
|
|
a*****g
发帖数: 19398 | 15 (转载)Why It's Hard to Create A Strong Go Software
For a long time it was a widely held opinion that computer Go posed a
problem fundamentally different to computer chess insofar as it was believed
that methods relying on fast global search compared to human experts
combined to relatively little domain knowledge would not be effective for Go
. Therefore, a large part of the computer Go development effort was during
these times focused on ways of representing human-like expert knowledge and
combini... 阅读全帖 |
|
w****i
发帖数: 964 | 16
word game:
MCT with heuristics is still heuristics, the most successful part of AlphaGo
is that it mimics human learning, which is building heuristics to
prioritize searching.
just go to kgs and look at darkfores1 and its related literature, it plays
instantly, unlike MCT always plays at the last second even no matter how
obvious the move is. |
|
O**l
发帖数: 12923 | 17
well
这个我当然知道 我们在讨论heuristic MCT是不是MCT的问题 和DCNN有什么关系， 我是
在强调MCT远远比heuristic重要 没MCT之前不管什么heuristic你1D都上不了 有了MCT
你能
上5D
pure
这是常识 DCNN拟合最多做做planning 你不search 简直开玩笑 你完全不理解DCNN MCT
的本质是啥 会不会下围棋也存疑
这是FB最新的paper里的话 你自己仔细读读再来说
http://arxiv.org/pdf/1511.06410v2.pdf
From the experiments, we clearly show that DCNN is tactically weak due to
the lack of search.
Search is a way to explore the solution space conditioned on the current
board situation, and build
a non-parametric local model for the game. The l... 阅读全帖 |
|
t******n
发帖数: 2939 | 18 ☆─────────────────────────────────────☆
Tianzi (tt) 于 (Tue Jan 31 08:31:13 2012, 美东) 提到:
忽悠的天才在国内太多了。记得以前读大学时，有一本必修课的书，作者两个，姓一样
。后来才知道，是父子。儿
子靠这本书申请美国的学校轻轻松松。书其实很烂。都是翻译的东抄西抄的东西。国内
这种事情太多，所以每当有个天才出来，我都不信。反正大多数的所谓天才，最后不是
被揭露，就是什么正经成就都没有，只有忽悠的成就。
国内每隔一段时间就会出现个天才，此起彼伏。包装，炒作天才已经形成产业链。但就
有那么多人民相信这个。体现了中国人民普遍缺乏最起码的critical thinking，就是
：惊世骇俗的claim，必须有惊世骇俗的证据。
☆─────────────────────────────────────☆
powell (lake) 于 (Tue Jan 31 08:40:15 2012, 美东) 提到:
天才就是说说，连学历，职称都不是。
☆──────────────────────────... 阅读全帖 |
|
w******d
发帖数: 1301 | 19 I said in my first post, I do not consider science is the ultimate truth,
while I do think science is important.
If you know about artificial intelligence, there is a so called A* algorithm
. One of the component is the local search, which is the expension of a open
, close table in the searching area. Another component is the heuristic
function that heuristically defines what is the final destination. The local
search might contradict against the heuristic function from time to time,
but it is |
|
s******e
发帖数: 493 | 20 The optimizer uses a mix of cost and heuristics to find a best plan for fast
delivery of the first row. Note: Using heuristics sometimes leads the CBO
to generate a plan with a cost that is significantly larger than the cost of
a plan without applying the heuristic. |
|
t*s
发帖数: 1504 | 21 machines have the ability to beat humans too
it's just that IBM didn't invest on go
it's not about complexity
go is more complex for machines, so is for humans
machines can always see through more steps. to beat huamns, they also need
better heuristic functions...
What Deep Blue did was...IBM collected thousands thousands of replays and
stored it in DB's memeory...you just need to do the same for go -- record
all the go games since the begining of time, use machine learning to derive
a good heur... 阅读全帖 |
|
t*s
发帖数: 1504 | 22 all computer-based game players use heuristic fuctions, unless the game can
be solved using brute force
i said, it's a NAIVE heuristic function. it's an example. nevertheless, such
a naive algorithm, if running on deep blue, can still beat many amateurs.
you need a better algorithm to beat the pros
your |
|
c******o
发帖数: 1277 | 23 同学，我论文就是关于这些算法的。
不是万用灵药。你一样要有具体对于围棋的optimization，哪那么容易。
还有的就是局部heuristic 对全局意义不大。全局heuristic 又回到原问题，很难做好
的。 |
|
v********e
发帖数: 1985 | 24 就是穷举＋heuristic pruning。所谓的deep learning就是产生heuristic pruning的
一种办法而已。 |
|
l***i
发帖数: 1309 | 25 For sophie and facebull, there are many people submitted successful solution
already, and the puzzle master themselves have their version of solutions.
Proof of NP-hardness does not stop people from looking for heuristics and
approximation solutions. sophie has a dynamic programming algorithm runs in
O(2^n*n^2) solution, which is way better than O(n!). facebull is more tricky
, and I don't see a dynamic programming solution, but some heuristic and
pruning should get you there. And those problems |
|
r*c
发帖数: 167 | 26 来自主题: JobHunting版 - 上一道小题 The ardendertat.com's code is not always correct. For an example, it will
return 121121121 as the next palin for input 120798387, where there exists a
smaller one: 120808021 .
The following code has no such shortcoming. The key idea is greedy +
heuristic.
#include
#include
#include
#include
#include
using namespace std;
class NextPalinSolution
{
public:
vector NextPalin(const vectorssrc){
vectorsrc(ssrc);
int ... 阅读全帖 |
|
n****1
发帖数: 77 | 27 找java高人做 java程序，有意请邮件w****[email protected]
兼职 这几天要有时间
Aims:
Gain deeper understanding of search algorithms and heuristics
Apply design patterns in the construction of an object-oriented program
Learn more about the Java class libraries
Correctness (automarked): 70%
Design, programming style and analysis of heuristic: 30% |
|
L*******t
发帖数: 782 | 28 这个heuristic难道就是传说中的定式？
heuristic |
|
O**l
发帖数: 12923 | 29 最后一句光靠 CNN 能到3D 我不信
MCT又不是纯monte-carlo是UCB tree guided monte-carlo, converge速度比传统
heuristic search高million倍
MCT把go的水平从30K提高到业余5D 绝对最大的突破 远比这次糅合CNN 突破大多了
MCT 后 机器9*9就无敌了 再次之前根本不敢想
almost
of
heuristics |
|
w****i
发帖数: 964 | 30
darkfores1 by facebook is kgs 2d, its rating chart is very close to 3d, ~2.
8d. it only started playing on kgs from late Nov 2015.
non-MCT GNUGO is ~ kgs 5~6K, UCB-MCT improved bots level from 5k to ~1-2d,
from 2d to 5~6d mainly rely on customized heuristics. DCNN heuristics
improved bots level from kgs 5~6d to 9d+
standard Go is a 19*19 game. |
|
O**l
发帖数: 12923 | 31 1 customized heuristics MCT任然是MCT
2 DCNN heuristics提高到9D还不确定
3 有论文关于专门DCNN的介绍吗 纯DCNN不计算能到3D开什么玩笑 DCNN和MCT根本不是
一个层面上的东西 没法代替 |
|
O**l
发帖数: 12923 | 32 搞笑 MCT以前基于各种knowledge的heuristic search多了去了 谁能达到1d水平
没有UCB你加再多knowledge都没有 所以重要MCT不是heuristic
DCNN下得快当然了 就是过一遍NN DCNN其实就是UCT展开时候prune branch
不用MCT
你这玩意想也能想出来 应该勺子莫名其妙的棋一大把
而且局部对杀 你不计算用DCNN softmax一个最大值？？？ 开玩笑吧
有paper拿来看看嘛
AlphaGo |
|
b****r
发帖数: 1272 | 33 不要故意弄两个名词上去就来滚球版砖家了
我问问你为什么pure enumeration就是100%的努力，山东人做得到吗？如果穷举有意志
力有革命精神就能做的话人家还设计什么heuristic，你是在故意侮辱山东人吗？
我来给你指出一下改进的方向，这个问题根本不能RELAX成LP，LP解根本没有意义。你
连upper bound都没有，既不知道自己enumeration能enumeration个啥来，又根本不知
道人家heuristic solution的gap，你比来比去在比毛啊。
我严重怀疑你用过CPLEX么，如果没有的话，去上一些MS PROGRAM的课，做两个PROJECT
。如果你老板没有CPLEX LICENSE的话，我老可以借你们用用。 |
|
f*1
发帖数: 837 | 34 One clarification of the 50-55 F temperature: Note that's the soil
temperature, not the air temperature. The soil temperature almost always
trail the air temperature. Crabgrass normally germinate when the soil
temperature reaches 55-60 F consecutively for a few days, so you need to
drop the pre-emergents before that happens, thus the 50-55 F guidance.
So how do you know when to drop? Either by heuristics or by checking the
actual soil temperature.
By heuristics, drop when Forsythia (连翘) is in... 阅读全帖 |
|
k**o
发帖数: 15334 | 35 你太落后了，现在杀毒早就不靠那个了。现在都是靠heuristic analysis engine，
分辨病毒木马特征进行杀毒。能绕过现有的heuristic analysis的木马病毒，
已经水平很不错了，当然popular了之后，确实会被测出来。但是14岁有这水平
的，我敢说比较天才了。 |
|
B*******e
发帖数: 3882 | 36 http://appleinsider.com/articles/12/12/07/uspto-invalidates-all
USPTO invalidates all claims of 'Steve Jobs' multi-touch patent
By Mikey Campbell
A decision published by the U.S. Patent and Trademark Office on Monday saw
an Apple patent, co-invented by late Apple cofounder Steve Jobs, covering
multi-touch functionality have all 20 of its claims invalidated after the
property was reexamined.
Source: FOSS Patents
The USPTO's review of Apple U.S. Patent No. 7,479,949 for a "Touch screen
device, met... 阅读全帖 |
|
f*****r
发帖数: 229 | 37 Yes, system guys always use heuristics. But I found that if you have very
clear understanding of theory, you will have more elegant and general
heuristic, and not only some tricks for your workload and dataset. Such as, if
you have deep understanding of queue theory, you can design better scheme,
even though for desiging a monitor to collect critical measured data. Also,
if you have good understanding to stochastic process and prediction, you can
have deep understanding how prediction works in |
|
S******o
发帖数: 111 | 38 Same here. You can check out the recent AAMAS proceedings. Traditionally, lots
ppl work on negotiation, coodination, etc with some heuristic approach. Now I
think ppl try to introduce more quantitive approach into this area.
Some other hot topics include: auction(combinatorial auctions etc), Game
theory (self-interested agents), etc. You can check out the webpages of some
major research groups in this area.
My feeling is that heuristic based approaches still dominate this area. |
|
n****y
发帖数: 106 | 39 ft..
When I checked some references, they all first say it is NP hard and they
use some heuristics to speed up.
I spent a whole day checking the details of a "patent" about this and found
the implementation is wrong.
It looks to me that it is a classical problem, and I am afraid some people
might did similar things before. I just need a trustable reference(simple,
correct and using good heuristics) and implement it.
The manager even asked to implement the exponential method(he thinks the
number |
|
g****u
发帖数: 252 | 40 这个要最优是NP难的。只能用heuristic。12306也只能使heuristic。
这个问题和赌约无关。等这次赌完大家要还有兴趣就再一起慢慢研究。
这次赌约就是赌一个抢票器能不能做出来。你们要说计数器也行。 |
|
g****u
发帖数: 252 | 41 车票调度算法和卖出率的事情上上星期已经讨论过了。goodbug也确认过
不是要解决NP难找最优算法。事实上老魏找满足要求的最小区间的
heuristic(不知道有没有实现）已经超过goodbug当时的要求了。
他zhaoce当时又不是不在线。现在程序都写出来一半了他再来提这个，
显然不是处于技术讨论的目的。
12306这样的规模要做NP难的最优化，这样性能的机群怕是世上还没有。
12306最多就是heuristic复杂点。要同时超越12306的性能和座位利用率
还是有肯能的。 |
|
g****u
发帖数: 252 | 42 你定义正确率那个帖子我没找到。
你说的两个版本，老魏做的是分配座位的版本，只不过用的是最简单的heuristic。
他给的协议是返回座位号的。
我有两个直觉：
1. 你说的简单版和复杂版，难度是一样的。如果要保证简单版返回的座位最后
复杂版都能分配出来，简单版必须在内部实现分配算法才行。(前提是不允许
中途换座。）
2. 不加别的任何domain knowledge，老魏的heuristic应该是近似最优的。
所谓的近似最优，是指在上下10倍时间开销范围内所可以实现的近似优化算法，
在优化效果上不会对老魏的算法有显著的改进。对于NPC问题，讨论理论最优值
没有多少实际意义。
我上次讨论专门提过，用户订票中的各种参数肯定不是均匀分布的。goodbug
回复说只要测均匀分布就行。实际每个路段的需求应该符合power law分布。
这个知识对优化策略是有影响的。如果一个路段很热门，你知道票最后肯定
能卖完的，优化时就可以少算点。一个路段很冷门，优化时可以多算点。
因为绝大多数请求都是热门路段，所以平均下来可以达到卖出率和速度兼顾。
这种东西如果要深究，本身就是个大坑。
实际座位分配铁道系统应... 阅读全帖 |
|
d*******r
发帖数: 3299 | 43 所以就是个 搜索+heuristic剪枝
只不过是用 deep learning 来做 heuristic剪枝? |
|
A***A
发帖数: 98 | 44 Looks like he is in the Kahneman-Tversky camp.
Any one on this board is in the Judgment/Decision Making area?
I know very little about JDM, but one of the things that keeps puzzling me is
that while the Kahneman-Tversky school of heuristics and biases offers
insights into the irrational part of ourselves, they almost never give a
normative model for such decision heuristics or biases, apart from
demonstrating the existence of such biases (I'm not sure if I'm correct on
this, so please explain to |
|
j**********i
发帖数: 3758 | 45 http://en.wikipedia.org/wiki/Heuristic
道路设计是经过许多人提醒，尤其是人口的分布信息的提醒极为重要。
道路设计还有的尾巴是城市道路设计，这涉及到城市规划问题。
这个问题有许多美国小孩参加，有一些小黑。这些小孩参与的是美国街道
长度和中国街道长度差别问题（差距明显）。小孩子提醒我这一点。
这很重要，中国城市设计师没有考虑过这个问题。
还有上下班时间的问题。这个问题和城市大小有相关。这需要不同人种的实验。
有人认为，上下班时间最多在30-45分钟为上限，这是根据人的体力，精力的
测算结果和观察。
要知道城市大小，紧接着的要问的是：步行速度，骑车速度，电动车速度，
私家车在非高速上的速度，地铁的速度。以及混合速度。
城市大小是高频交通问题，以前我分析失误过，被人指出。
高频问题的积累是可观的，这是根据人口测算，居住分布和上班地点，以及经验
得出的结论。以前，为了算出租车搭载问题，碰过不可逾越的墙。所以，知道
这个问题的重要性。中国没有低频长距离物流和高频短距离交通的油耗对比。
最后，地铁拉的过长产生的不确定性很可怕，在南京这里已经有显现，城市
设计师的... 阅读全帖 |
|
c******r
发帖数: 300 | 46 我不否认CS的人确实在写产品上贡献很多(这些是为啥相对统计pay的好的道理)，可是
你要说理论部分CS对data science的贡献摆脱你多读点paper，很多original idea都是
数学统计的人做的(NN, PCA, discriminant analysis, kernel methods, EM，
graphical model, decision tree, model selection, regularization/shrinkage,
this list goes on ...), CS只是做了很多heuristics, adaptation，improvement
and make it HOT，POPULAR and ALMOST FAD，但是这不代表这些东西是CS发明的吧。
不过话说回来你如果只看现在的paper当然不会知道这些了，就像今天谁还去cite这些
最开始的paper。另外做统计的人确实比较不喜欢heuristic的东西，所以比较跟不上潮
流，搞得自己本来有的credit都没了
,
此,
处.
non- |
|
z****k
发帖数: 1057 | 47 你觉得复姓人少，是因为你觉得单姓人多
你觉得单姓人多，是因为你可以很容易想到很多单姓有许多人的例子，例如李王张刘
或者说，提到单姓就能够想到它们（李王张刘），而不是另一些生僻单姓，
是因为你可以想到很多姓李王张刘的人。这个叫availability heuristic.
因此，事实上你做出“复姓人少单姓人多”的结论，
是基于比较*大*单姓和复姓的结果,
而非比单纯比较单姓和复姓的结果。 |
|
|
f**k
发帖数: 15238 | 49 elaborate on heuristic please.
derive |
|
