R*******n
发帖数: 428 | 1 到MITBBS，我主要看两个版，羽毛球，围棋。虽然数学是我的专业，
但很少去数学版，几乎没在那里发个贴。昨天发了一个贴，今天发
现居然上了首页。收到版主来函说奖励伪币20。到现在仍然不知道
这个伪币有什么用？
数学版那个帖子叫“智力上限”，换成羽毛球语言就是，99.99%的
人即使从出生就开始接受专业训练，也达不到林丹的水平。 | e****f
发帖数: 3393 | 2 伪币就是买买提玩家家用的。相当于玩牌时候的筹码。
蛋神那种境界，是需要天才加勤奋的，缺一不可。我们大部分人没那天分！但是并不妨
碍我们对快乐的追求。
【在 R*******n 的大作中提到】
: 到MITBBS，我主要看两个版，羽毛球，围棋。虽然数学是我的专业，
: 但很少去数学版，几乎没在那里发个贴。昨天发了一个贴，今天发
: 现居然上了首页。收到版主来函说奖励伪币20。到现在仍然不知道
: 这个伪币有什么用？
: 数学版那个帖子叫“智力上限”，换成羽毛球语言就是，99.99%的
: 人即使从出生就开始接受专业训练，也达不到林丹的水平。
| l*********y
发帖数: 3447 | 3 I disagree with what you said. There are difference among professional
training, there are difference among the life track everyone experience,
from born to death. And those details make big difference. Unless you have
people grow up in exactly the very same condition lin dan experienced, you
could not possibly prove your theory.
The "resolution" of training and individual experience is way higher than
that of the level of playing badminton, or doing math.
【在 R*******n 的大作中提到】
: 到MITBBS，我主要看两个版，羽毛球，围棋。虽然数学是我的专业，
: 但很少去数学版，几乎没在那里发个贴。昨天发了一个贴，今天发
: 现居然上了首页。收到版主来函说奖励伪币20。到现在仍然不知道
: 这个伪币有什么用？
: 数学版那个帖子叫“智力上限”，换成羽毛球语言就是，99.99%的
: 人即使从出生就开始接受专业训练，也达不到林丹的水平。
| m****r
发帖数: 6639 | 4 数学的theory和自然、社会科学的theory是不一样的东西。 科学界的theory拿到数学
里面都是hypothesis。 拿到计算机里面都是np hard。 基本不可能prove它正确，
只能找反例。
【在 l*********y 的大作中提到】
: I disagree with what you said. There are difference among professional
: training, there are difference among the life track everyone experience,
: from born to death. And those details make big difference. Unless you have
: people grow up in exactly the very same condition lin dan experienced, you
: could not possibly prove your theory.
: The "resolution" of training and individual experience is way higher than
: that of the level of playing badminton, or doing math.
| k*w
发帖数: 2063 | 5 概率论就是研究这些不确定性的东东。
【在 m****r 的大作中提到】
: 数学的theory和自然、社会科学的theory是不一样的东西。 科学界的theory拿到数学
: 里面都是hypothesis。 拿到计算机里面都是np hard。 基本不可能prove它正确，
: 只能找反例。
| R*******n
发帖数: 428 | 6 > I disagree with what you said. There are difference among
> professional training, there are difference among the life
> track everyone experience, from born to death. And those
> details make big difference. Unless you have people grow
> up in exactly the very same condition lin dan experienced,
> you could not possibly prove your theory. The "resolution"
> of training and individual experience is way higher than
> that of the level of playing badminton, or doing math.
现在的世界跳高记录是2米45。如果我说一个身高1米45的人，无论如何训练都
不可能跳那么高，你应该没有意见。这是因为很明显的先天高度差距摆在那里。
其它方面的先天差距（比如羽毛球）就不是那么明显了。不管明显不明显，我认
为普通人与各个项目的世界冠军的差距都是差不多的。跑步也好，游泳也好，下
围棋也好，最后达到顶尖地位的人都是在一大堆同样刻苦训练的孩子中一层一层
地筛选出来的。他们比别人多的就是天赋（在同样努力的基础上）。我的论断就
是基于这样一条公理：“任何项目拿过世界冠军的人，都具有常人没有的天赋，
而这种天赋不是可以通过刻苦训练可以弥补的”。如果你不接受这条公理，我们
就没有讨论下去的必要，因为体系不同。公说公有理，婆说婆有理，永远争不出
结果。如果你接受这条公理，那么，我的结论就很显然了。林丹是世界冠军，他
身上有常人没有的天赋。这个公理也可以推广到智力上。数学上没有世界冠军，
但有大家公认的领军人物。这些人就有常人没有的天赋。如果你在大学教过非数
学专业的微积分，你就有很多切身体会，很多人真是数学上缺根筋，无论如何都
搞不懂。更不要说达到世界顶尖地位了。 | l*********y
发帖数: 3447 | 7 不接受这条公理
【在 R*******n 的大作中提到】
: > I disagree with what you said. There are difference among
: > professional training, there are difference among the life
: > track everyone experience, from born to death. And those
: > details make big difference. Unless you have people grow
: > up in exactly the very same condition lin dan experienced,
: > you could not possibly prove your theory. The "resolution"
: > of training and individual experience is way higher than
: > that of the level of playing badminton, or doing math.
: 现在的世界跳高记录是2米45。如果我说一个身高1米45的人，无论如何训练都
: 不可能跳那么高，你应该没有意见。这是因为很明显的先天高度差距摆在那里。
| y****n
发帖数: 782 | 8 看来楼上那位就是属于你所说的缺根筋的范畴里了。
【在 R*******n 的大作中提到】
: > I disagree with what you said. There are difference among
: > professional training, there are difference among the life
: > track everyone experience, from born to death. And those
: > details make big difference. Unless you have people grow
: > up in exactly the very same condition lin dan experienced,
: > you could not possibly prove your theory. The "resolution"
: > of training and individual experience is way higher than
: > that of the level of playing badminton, or doing math.
: 现在的世界跳高记录是2米45。如果我说一个身高1米45的人，无论如何训练都
: 不可能跳那么高，你应该没有意见。这是因为很明显的先天高度差距摆在那里。
| m****r
发帖数: 6639 | 9 我也不接受你的公里。
天赋这个东西明显存在。 你写得不太清楚的是， 它是不是可以被更细的分类， 量化
。 拿打羽毛球来说， 手感， 弹跳， 球路， 体力， 这些都可以分开说。 比如，
你自己体力上的天赋就很好。 我体力有一点点天赋， 但是可以没有你好。 然后有
些人完全没有体力上的天赋。
所以， 天赋和grace应该不一样。 （you either have grace, or you don't have
grace. you can't just have a little bit of grace).
然后， 说世界冠军。 我觉得很多世界冠军都没有天赋。 当然， 也有很多有很高的
天赋。
【在 R*******n 的大作中提到】
: > I disagree with what you said. There are difference among
: > professional training, there are difference among the life
: > track everyone experience, from born to death. And those
: > details make big difference. Unless you have people grow
: > up in exactly the very same condition lin dan experienced,
: > you could not possibly prove your theory. The "resolution"
: > of training and individual experience is way higher than
: > that of the level of playing badminton, or doing math.
: 现在的世界跳高记录是2米45。如果我说一个身高1米45的人，无论如何训练都
: 不可能跳那么高，你应该没有意见。这是因为很明显的先天高度差距摆在那里。
| R*******n
发帖数: 428 | 10 > 我也不接受你的公里。
> 天赋这个东西明显存在。 你写得不太清楚的是， 它是不是可以被更细
> 的分类， 量化。 拿打羽毛球来说， 手感， 弹跳， 球路， 体力， 这
> 些都可以分开说。
你说的都有道理，我们可以继续讨论。
我说的天赋当然是全面的综合素质。林丹速度不是最好的，弹跳不是最好的，
爆发力也不是最好的。还有其它许多羽毛球需要的身体和智力因素他或许都
不是最好的。但是，羽毛球需要的所有因素综合起来，马上就把他放到6个
方差以外（70亿分之5），成了全世界屈指可数的Outlier。普通大众(1,2
个方差，甚至3个方差之内的人) 无论如何都跑不到他那里去。我那条公理
是有一些统计背景的。:)
当然，这些东西不像数学，没有绝对的对错。能被说服的人都必须要接受一
定的假设。比如我前面提到的，如果你承认身高1米45的人训练一辈子不能
跳过2米45，你就要承认在羽毛球综合能力高度1.45的人训练一辈子达不到
2.45。我们的区别在于，你（还有Lovecountry）不承认有羽毛球综合能力
这个虚拟高度。这方面我没有办法说服你们。:) | | | m****r
发帖数: 6639 | 11 我没有说我不接受有这个综合素质呀。 我是说你自己对你的公里的定义没有把这个综合
素质说清楚。
比如， 一定要6个方差之外的才叫天赋吗？ 还是5个也可以算。 还是几个？
然后， 你的公里的原话是“世界冠军肯定有天赋”。 这个我还是不能接受。 上次加
拿大有个剪刀石头布的世界竞标赛， 也出了一个世界冠军。
【在 R*******n 的大作中提到】
: > 我也不接受你的公里。
: > 天赋这个东西明显存在。 你写得不太清楚的是， 它是不是可以被更细
: > 的分类， 量化。 拿打羽毛球来说， 手感， 弹跳， 球路， 体力， 这
: > 些都可以分开说。
: 你说的都有道理，我们可以继续讨论。
: 我说的天赋当然是全面的综合素质。林丹速度不是最好的，弹跳不是最好的，
: 爆发力也不是最好的。还有其它许多羽毛球需要的身体和智力因素他或许都
: 不是最好的。但是，羽毛球需要的所有因素综合起来，马上就把他放到6个
: 方差以外（70亿分之5），成了全世界屈指可数的Outlier。普通大众(1,2
: 个方差，甚至3个方差之内的人) 无论如何都跑不到他那里去。我那条公理
| R*******n
发帖数: 428 | 12 > 一定要6个方差之外的才叫天赋吗？ 还是5个也可以算。 还是几个？
5个方差以外全世界有2千多人，上面一半有1000+，太多。
> 然后， 你的公里的原话是“世界冠军肯定有天赋”。 这个我还是不能接受。 上次加
> 拿大有个剪刀石头布的世界竞标赛， 也出了一个世界冠军。
剪刀石头布的世界竞标赛我没想到，确实应该定义清楚一点。比如说；“有20个国家以上
参加的奥运项目的世界冠军”。 | m****r
发帖数: 6639 | 13 so, that is 6 standard deviations above the mean, as the definition of "tain
fu".
that alone, is enough, right? why do you need to add all world champions?
i *think* most people would say that xia xuanze, zhang ning, wang yihan, are
all players with not so much gift but trained really hard.
i would also name all french open winners whose name is not nadal as
giftless.
anyway, changes of a trely gifted person to train and compete in the sport
that he/she is gifted in is still very small. and with the gifted people
that is left in the field, lack of motivation, injury, social/economical
issues, would often render them not as great as they could have been. so,
it is hard for me to believe that the small field of gifted people will
always win all th world championships.
以上
【在 R*******n 的大作中提到】
: > 一定要6个方差之外的才叫天赋吗？ 还是5个也可以算。 还是几个？
: 5个方差以外全世界有2千多人，上面一半有1000+，太多。
: > 然后， 你的公里的原话是“世界冠军肯定有天赋”。 这个我还是不能接受。 上次加
: > 拿大有个剪刀石头布的世界竞标赛， 也出了一个世界冠军。
: 剪刀石头布的世界竞标赛我没想到，确实应该定义清楚一点。比如说；“有20个国家以上
: 参加的奥运项目的世界冠军”。
| R*******n
发帖数: 428 | 14 > i *think* most people would say that xia xuanze, zhang ning,
> wang yihan, are all players with not so much gift but trained
> really hard.i would also name all french open winners whose
> name is not nadal as giftless.
My original claim was "ordinary people cannot be trained to reach Lin Dan's
level". I gave two definition of ordinary people, the first one was 99.99%,
the second one was 1 or 2 standard deviation, may be even 3 standard
definition. The two definition are compatible.All the people you listed are
in the 5 std range, or at least 4.5 range. After all, the total range should
be an almost continuous curve. People within 1 std away may be reachable
through hard training (may be "twice" as hard). People who's 2 std away have
to train three time as hard, and 3 std away need to train 4 times as hard.
And there may be no such thing as 4 times as hard as Lin Dan, if Lin Dan is
already training as almost the top level hard. :) | m****r
发帖数: 6639 | 15 your original axiom was "all world champions have extraordinary gift which
training cannot compensate".
I asked you to define "extraordinary", you said 6 std above mean.
so, i see inconsistency here. either the definition of extraordinary needs
to be changed, or the std estimates for the people I listed needs to be
upgraded, or your axiom is in fact faulty.
right? you su su?
s
,
are
should
have
.
is
【在 R*******n 的大作中提到】
: > i *think* most people would say that xia xuanze, zhang ning,
: > wang yihan, are all players with not so much gift but trained
: > really hard.i would also name all french open winners whose
: > name is not nadal as giftless.
: My original claim was "ordinary people cannot be trained to reach Lin Dan's
: level". I gave two definition of ordinary people, the first one was 99.99%,
: the second one was 1 or 2 standard deviation, may be even 3 standard
: definition. The two definition are compatible.All the people you listed are
: in the 5 std range, or at least 4.5 range. After all, the total range should
: be an almost continuous curve. People within 1 std away may be reachable
| k*w
发帖数: 2063 | 16 呵呵，我插一句：“幸福的家庭都是相似的 不幸的家庭各有各的不幸”。
最后的成就，是受到最弱的一个环节限制。所以1.45m的人跳不了2.45m. 无论你如何努
力。
needs
【在 m****r 的大作中提到】
: your original axiom was "all world champions have extraordinary gift which
: training cannot compensate".
: I asked you to define "extraordinary", you said 6 std above mean.
: so, i see inconsistency here. either the definition of extraordinary needs
: to be changed, or the std estimates for the people I listed needs to be
: upgraded, or your axiom is in fact faulty.
: right? you su su?
:
: s
: ,
| R*******n
发帖数: 428 | 17 > your original axiom was "all world champions have extraordinary gift
> which training cannot compensate".
This should read as "all world champions have extraordinary gift
which 'ordinary' people through training cannot compensate"
I always assumed we are talking about 'ordinary' people (i.e. people within
1,2,3 std of the mean) vs World Champions. As I said before, within 1 std (
eg. 5 vs 6) is reachable through training.
You are right, I need to be more clear about my definitions, not to depend
on implicit assumptions.
I am glad we had these discussions, it makes everything more clear. | m****r
发帖数: 6639 | 18 ok.
so, we agree that there is a compensatory relationship between gift and
training.
i think it is not interesting to say that no matter how hard ordinary people
train, they cannot be as good as a hard training gifted person. after all,
it is very hard to find other players who train as hard as lin dan has done.
I am more interested in finding the formula that translates gift into years
of training. ie, assuming A is an average person and B is 1 std above the
mean when it comes to gift, how many years does A have to train to just as
good as B? if B has had x years of training.
within
【在 R*******n 的大作中提到】
: > your original axiom was "all world champions have extraordinary gift
: > which training cannot compensate".
: This should read as "all world champions have extraordinary gift
: which 'ordinary' people through training cannot compensate"
: I always assumed we are talking about 'ordinary' people (i.e. people within
: 1,2,3 std of the mean) vs World Champions. As I said before, within 1 std (
: eg. 5 vs 6) is reachable through training.
: You are right, I need to be more clear about my definitions, not to depend
: on implicit assumptions.
: I am glad we had these discussions, it makes everything more clear.
| R*******n
发帖数: 428 | 19 > I am more interested in finding the formula that translates
> gift into years of training. ie, assuming A is an average
> person and B is 1 std above the mean when it comes to gift,
> how many years does A have to train to just as good as B?
> if B has had x years of training.
The whole discussion is about whether such a formula exists or not.
My claim is that the relationship between training and achievement
is not linear, and everyone has a plateau (i.e. the highest level
they can reach, no matter how hard they train), using std as unit,
it is more like
"achieve-level" = h*"build-in-level" + g*(1-exp(-k*training))
where, build-in-level is that person's gift in terms of std,
and k is a constant. h may be a constant associated with basic
training, something like h = min(1,training). So that, it is
a constant after one year of training. This is introduced so
that a highly gifted person still needs some basic training.
If this formula is correct, then, a person A who's g unit behind
another person B, then, A will never be able to reach B no matter
how much he trains.
You can change the model for h. You can change the coefficient k,
you can definitely change g to some other numbers.But my claim
is that number should not be very far from 2.
Sorry to use math terms here, but your reply prompted all of this. :) | m****r
发帖数: 6639 | 20 there is no way for lin dan to be as good as you or me with just one year of
training. of course, i have to be careful with the definition of training.
if he was training for track and field for 10 years and then gets one year
of badminton training, that is not just 1 year of training. that is 11
years of training because physical training obviously matters.
and for you and me, although we are not "training", all the recreational
games we play every week, account for something.
my formula is closer to something like:
L = k * BIL * total-training
so, basically, for each year lin dan trains, because he's gifted, he gets
more better than the average person. and because he never stops training,
nobody can catch up.
i think the plateau you refer to exists, but that is more closely related to
the physical limits of each human body. but training always make you
better, no matter what.
【在 R*******n 的大作中提到】
: > I am more interested in finding the formula that translates
: > gift into years of training. ie, assuming A is an average
: > person and B is 1 std above the mean when it comes to gift,
: > how many years does A have to train to just as good as B?
: > if B has had x years of training.
: The whole discussion is about whether such a formula exists or not.
: My claim is that the relationship between training and achievement
: is not linear, and everyone has a plateau (i.e. the highest level
: they can reach, no matter how hard they train), using std as unit,
: it is more like
| | | R*******n
发帖数: 428 | 21 > there is no way for lin dan to be as good as you or me
> with just one year of training.
As I said in my previous post, you can certainly change the definition of h,
may be it is not 1 year, may be it needs x year to get to basic
professioncy (I just made up this word).
> my formula is closer to something like:
> L = k * BIL * total-training
the problem with this model I see is that the latter training has the same
effect as earlier training, which is almost definitely not true. We all know
that the first several year of training pays much more than latter training
. e.g. one can improve the level of WeiQi a lot in the first several years,
and the improvement becomes slower and slower. Mathemtically speaking, it is
plateaued.
I find these discussions very helpful, thanks. | m****r
发帖数: 6639 | 22 with regard to the linearality of the outcome of training, we never
discussed how to measure the outcome, so, there is no basis to discuss what
pays more.
in weiqi, you could potentically measure it by playing against the same
computer program and measure the result of how much you win or lose. while
that is not necessarily a good way to measure "level", it is A way.
if we were to use this way to measure weiqi level, I am not convince latter
training could result in slower gain of level. your winning margin could
jump dramtically.
and the same could be imagined for badminton. we build a robot, and
configure it with a certain speed, power, accuracy. and play against it and
see how many points you can win or lose. against, in this case, i am not
convinced latter training result in slower gain.
h,
know
training
,
is
【在 R*******n 的大作中提到】
: > there is no way for lin dan to be as good as you or me
: > with just one year of training.
: As I said in my previous post, you can certainly change the definition of h,
: may be it is not 1 year, may be it needs x year to get to basic
: professioncy (I just made up this word).
: > my formula is closer to something like:
: > L = k * BIL * total-training
: the problem with this model I see is that the latter training has the same
: effect as earlier training, which is almost definitely not true. We all know
: that the first several year of training pays much more than latter training
| t*******e
发帖数: 723 | 23 很有趣的讨论，我居然一路看下来了。
建议世界羽联采用你的公式给运动员排名，
当然应该在加上一项 比赛积分，这样似乎更
公平。两个常数可以通过收集所有运动员
材料率定出来，省得这帮人整天不务正业，
哈哈。
【在 R*******n 的大作中提到】
: > I am more interested in finding the formula that translates
: > gift into years of training. ie, assuming A is an average
: > person and B is 1 std above the mean when it comes to gift,
: > how many years does A have to train to just as good as B?
: > if B has had x years of training.
: The whole discussion is about whether such a formula exists or not.
: My claim is that the relationship between training and achievement
: is not linear, and everyone has a plateau (i.e. the highest level
: they can reach, no matter how hard they train), using std as unit,
: it is more like
| R*******n
发帖数: 428 | 24 > I am not convince latter training could result in slower
> gain of level.
I thought it is a well known fact that people improve the fastest at the
beginning of any learning. Obviously, it is not as well accepted as I
thought. :) Very interesting.
> in weiqi, you could potentically measure it by playing against
> the same computer program and measure the result of how much
> you win or lose. while that is not necessarily a good way to
> measure "level", it is A way.if we were to use this way to measure
> weiqi level, your winning margin could jump dramtically.
OK, we will use your WeiQi Robert as the model.
Assume the WeiQi robot is at amateur 5 dan level (this is around the current
highest level of the WeiQi program). At the beginning period,it makes no
sense to measure the winning ratio of a beginner vs the robot, because, s/he
will loss every time, i.e. 100% for the first year. However, there's
another measure we can take. We can count how many points s/he loses. At the
beginning, he lose 300 points. After 1 month, s/he loses 200 points. After
another 4 month, he loses 100 points. After another year, he loses 50 points
. Two more year, he loses 25 points. The gain is getting smaller and smaller
. For most people (i.e. more than 50%), they will never be able to shrink
the gap down to zero. Some people may be able to eventually beat the robot.
And the gap starts to get bigger in another direction. But, there will be no
time that s/he can increase the gap by 100 points after one more month of
training. The bottom line is, I think there is no measure that
can validate your linear growth training-pay formula forever.There must be a
plateau. | k*w
发帖数: 2063 | 25 赞一路看下来了。
【在 t*******e 的大作中提到】
: 很有趣的讨论，我居然一路看下来了。
: 建议世界羽联采用你的公式给运动员排名，
: 当然应该在加上一项 比赛积分，这样似乎更
: 公平。两个常数可以通过收集所有运动员
: 材料率定出来，省得这帮人整天不务正业，
: 哈哈。
| m****r
发帖数: 6639 | 26 I totally ft. of course what i meant was measuring how many points you win
or lose.
current
he
the
After
【在 R*******n 的大作中提到】
: > I am not convince latter training could result in slower
: > gain of level.
: I thought it is a well known fact that people improve the fastest at the
: beginning of any learning. Obviously, it is not as well accepted as I
: thought. :) Very interesting.
: > in weiqi, you could potentically measure it by playing against
: > the same computer program and measure the result of how much
: > you win or lose. while that is not necessarily a good way to
: > measure "level", it is A way.if we were to use this way to measure
: > weiqi level, your winning margin could jump dramtically.
| m****r
发帖数: 6639 | 27 ok, continue to use the weiqi for case study. i don't see why using the
highest level robot is the best way to measure this.
one could argue using the robot whose level is at mean is more fair?
and if there was a 9d robot, your first 3 years of training would result in
almost no gain? that would also dis-prove the assumption that earliest
learning results in more gain.
at the end of the day, there is NO measurement for level. and it is very
hard to put the concept of "level" in numeric values, which is the
prerequisite for discussing linearality or speed or acceleration.
for example, you cannot really say "the ammount of increased level of a 5d
compared to a 4d is the same as a 4d compared to a 3d". it looks linear in
terms of Dan, but it says nothing of the actual level of play, since there
is no numeric measurement.
in a sport like track, you could potentially say that it is alot harder to
go from 9.4 to 9.3 compare to going from 13.5 to 13.4. but I would argue
that you are basically the same athelete after going from 13.5 to 13.4, but
you have just went from being amazing to godlike in the other case. same .1
seconds, totally different gain in level.
current
he
the
After
【在 R*******n 的大作中提到】
: > I am not convince latter training could result in slower
: > gain of level.
: I thought it is a well known fact that people improve the fastest at the
: beginning of any learning. Obviously, it is not as well accepted as I
: thought. :) Very interesting.
: > in weiqi, you could potentically measure it by playing against
: > the same computer program and measure the result of how much
: > you win or lose. while that is not necessarily a good way to
: > measure "level", it is A way.if we were to use this way to measure
: > weiqi level, your winning margin could jump dramtically.
| R*******n
发帖数: 428 | 28 > it looks linear in terms of Dan, but it says nothing of the actual
> level of play, since there is no numeric measurement.
I proposed a numeric measurement, the number of points lost in a game, but
you choose to ignore it, and claim there is no numeric measurement.:)
Actually, many Korean weiqi players use this system when they gamble. 1
point equal to $10. I watched one guy lose $300 in one night (3 games).
For badminton, if I play with Lin Dan, I will probably get 1 point in a game
by hitting the bird to the net and tumbling over. Whereas, Sattwat, lost to
Lin Dan 21:6 (in two games). He's definitely better than me. :) What I want
to say is, points in a game is a good measurement for skill.
If you want to compare anything, you have to turn it into numbers. People
used to say that we cannot compare apple and orange. But if you want to
compare the sweetness of the two, there is definitely a numeric way to do it
.
OK, I will give you another "numeric" way to compare learning-pay relation.
An average kid get badminton training for one month, he will be definitely 1
std above the mean level of badminton play among general population. Two
more months of training, he will be 2 std above the mean. One year of
training, he will be 3 std above the mean. And the amount of std unit he
gain is getting smaller and smaller for each month (or year) or training.
Most people will not be able to get to 5 std above the mean no matter how
hard they train. The std unit can turn into number of people they pass after
training. Isn't this a good numeric approach? | m****r
发帖数: 6639 | 29 you need to read what I said again. I didn't ignore your proposal. I am
questioning you why you choose 5D weiqi or LinDan for the "standard". why
is that better than choosing a 4k robot or your average B player for the "
standard"?
so, you are right, points in a game is a good measurement. but points in a
game against who?
and to your std proposal, it doesn't make sense. to get std, you must first
have unprotected sex, or have a numeric system already. you cannot propose
the first numeric system based on a non-existing numeric system.
game
to
want
【在 R*******n 的大作中提到】
: > it looks linear in terms of Dan, but it says nothing of the actual
: > level of play, since there is no numeric measurement.
: I proposed a numeric measurement, the number of points lost in a game, but
: you choose to ignore it, and claim there is no numeric measurement.:)
: Actually, many Korean weiqi players use this system when they gamble. 1
: point equal to $10. I watched one guy lose $300 in one night (3 games).
: For badminton, if I play with Lin Dan, I will probably get 1 point in a game
: by hitting the bird to the net and tumbling over. Whereas, Sattwat, lost to
: Lin Dan 21:6 (in two games). He's definitely better than me. :) What I want
: to say is, points in a game is a good measurement for skill.
| m****r
发帖数: 6639 | 30 another reason why std proposal doesn't make sense, is because it only
confirms the assumption that player level is normally distributed. (
therefore, it gets harder and harder to move away from the mean). it says
nothing about the hardship to gaining higher level.
you tried to be tricky. but i caught you.
game
to
want
【在 R*******n 的大作中提到】
: > it looks linear in terms of Dan, but it says nothing of the actual
: > level of play, since there is no numeric measurement.
: I proposed a numeric measurement, the number of points lost in a game, but
: you choose to ignore it, and claim there is no numeric measurement.:)
: Actually, many Korean weiqi players use this system when they gamble. 1
: point equal to $10. I watched one guy lose $300 in one night (3 games).
: For badminton, if I play with Lin Dan, I will probably get 1 point in a game
: by hitting the bird to the net and tumbling over. Whereas, Sattwat, lost to
: Lin Dan 21:6 (in two games). He's definitely better than me. :) What I want
: to say is, points in a game is a good measurement for skill.
| | | l*********y
发帖数: 3447 | 31 ...what took you so long?
【在 m****r 的大作中提到】
: another reason why std proposal doesn't make sense, is because it only
: confirms the assumption that player level is normally distributed. (
: therefore, it gets harder and harder to move away from the mean). it says
: nothing about the hardship to gaining higher level.
: you tried to be tricky. but i caught you.
:
: game
: to
: want
| R*******n
发帖数: 428 | 32 > you tried to be tricky. but i caught you.
I am seriously discuss this with you, not trying to be tricky. I am writing
a long article about this, and this discussion makes my writing more clear.
Thanks again.
> I am questioning you why you choose 5D weiqi or LinDan for the
> "standard". why is that better than choosing a 4k robot or your
> average B player for the "standard"?
Ok, we use your approach, we have a series of WeiQi robot, ranking from 20k,
19k, 18k,....,1k, 1dan, 2dan,...,9dan
At the beginning, a player will find that every month training will pay off,
he could fly by those k's one after another. Then, he slows down. Not that
he slows down training, he actually find the game more and more interesting,
and spend all his time on it, but each new robot takes longer and longer
time to conquer. The pay for the training is not as rapid and noticeable as
before. And it is getting harder and harder. I remember when I was
programing intelligent game software 20+ years ago (for fun, not for work),
the difference between different level (that user can choose) are simply the
degree of search level. That may very well be the difference between those
robots of different ranking, a very concrete down to earth numerical
measurement, i.e. the number of search steps (recursive) that the program
uses. As the search level increases, each additional search level
corresponds to longer and longer training time, while it is the same
increase pattern for the computer.
> and to your std proposal, it doesn't make sense. to get std, you
> must first have unprotected sex, or have a numeric system already.
> you cannot propose the first numeric system based on a non-existing
> numeric system.
Let's get the terminology straight, std = standard deviation.
And std is a very verifiable numerical system. Hypothetically speaking,
after each month of training, we can go on the street and grab a bunch of
average Joe's to play against him, and see how he does against them. When he
gets to 3 or 4 std level, we will ask him to play against the city champion
, the state champion, then, the US champion, then, the world top 20, and so
on. Each level is very verifiable.
Thanks again. | m****r
发帖数: 6639 | 33 there are so many different ideas in your post. i am going to respond one
after another, in different posts.
using different robots is not my approach. my approach is to use one robot,
just like your own approach. the difference is, you are using a 5D robot.
I am using a 5k robot. if we use my approach, it may very well be the case
that the ammount of points you win can increase dramatically after a
certain level, and that translate into latter training produces more gain.
I want to emphasize that the only difference between your approach and my
approach is the level of the robot.
and in response to the 20k .. to .. 9d robot approach, it is the same point
i am trying to make for the whole time, which you don't seem to try to
understand. the K and D system is an attempt to map the concept of "level"
to numbers. In badminton, we try to do it with letters, A/B/C/D. but there
is no evidence to support that these mappings are actually linear. more on
this in my next post, where I respond to the AI search tree.
writing
.
20k,
off,
that
interesting,
as
【在 R*******n 的大作中提到】
: > you tried to be tricky. but i caught you.
: I am seriously discuss this with you, not trying to be tricky. I am writing
: a long article about this, and this discussion makes my writing more clear.
: Thanks again.
: > I am questioning you why you choose 5D weiqi or LinDan for the
: > "standard". why is that better than choosing a 4k robot or your
: > average B player for the "standard"?
: Ok, we use your approach, we have a series of WeiQi robot, ranking from 20k,
: 19k, 18k,....,1k, 1dan, 2dan,...,9dan
: At the beginning, a player will find that every month training will pay off,
| m****r
发帖数: 6639 | 34 I am glad you brought up the search tree, I think it is the example I am
looking for.
In the search tree example, to gain a new level, we have to put in extra
resources that is not linear to the number of levels.
So, I want to change the terminology here. we will have tree level, and
strength (and discontinue the use of "level" because it will now be
confusing).
the question is, is the tree level is good mapping of the the strength of
the program. Obviously, yes. The more levels you can search, the stonger
the program, no arguement there. and obviously, the resources needed is not
linear to the tree level, it is in fact exponential.
so, using the tree level to map the strength of the program is certainly
reasonable.
but, I can also use a different mappings to indicate strength, some of these
are: cpu time, nodes checked, memory consumed, etc. with everything else
being equal, the having more of each of these things, will always produce a
stronger outcome. right?
so, if you produce a 10 tree level program, I can produce a 10 1/n th level
program that will always be slightly better than yours. that fact that in
your ranking system there is no such thing as 10 1/n, doesn't mean it doesn'
t exist.
and put it in weiqi terms, there is a strength that is slightly stronger
than 5d and alot worse than 6d. the K/D system just choose to ignore it for
simplicity. but it actually does exist.
writing
.
20k,
off,
that
interesting,
as
【在 R*******n 的大作中提到】
: > you tried to be tricky. but i caught you.
: I am seriously discuss this with you, not trying to be tricky. I am writing
: a long article about this, and this discussion makes my writing more clear.
: Thanks again.
: > I am questioning you why you choose 5D weiqi or LinDan for the
: > "standard". why is that better than choosing a 4k robot or your
: > average B player for the "standard"?
: Ok, we use your approach, we have a series of WeiQi robot, ranking from 20k,
: 19k, 18k,....,1k, 1dan, 2dan,...,9dan
: At the beginning, a player will find that every month training will pay off,
| R*******n
发帖数: 428 | 35 > I want to emphasize that the only difference between your
> approach and my approach is the level of the robot.
I am glad that you see the "difference". The dispute we have is about
whether the train-pay relation is linear or not. You choose the level at
which linear is still a good approximation. Whereas,I choose a level when
the plateau effect takes place. Your approach has its merit, and I don't
dispute that. After all, everything can be linearized locally to a good
degree, but not globally. My claim is that the linear behavior will not
continue, and it will plateau at high level, i.e. Lin Dan's level, or 5+ Dan
level. You just choose to stick to the beginning portion of the curve,
which is never my intention to discuss.
> The more levels you can search, the stronger, the program, no
> argument there. and obviously, the resources needed is not
Good, we have finally found a common ground, "no argument there". And yes,
the higher level it gets, the more resources it is needed, and it is not
linear. That is the whole point that I am trying to make. You cannot use a
linear model to model a non-linear process. And, also, my original claim (
back in the very first post), most people don't have the resources (mentally
, physically) to reach to high level.
I am really enjoying the discussion. | x***8
发帖数: 369 | 36 原以为崇拜世界冠军就是和他们照照相而已...
老吉看到公理该偷偷乐了：最牛的天赋就是好运气！ |
|
