s***1 发帖数: 49 | 1 如果有一组数据,然后分别用 linear kernel (Xi dot Xj) , quadratic kernel ( (
Xi dot Xj+1)^2 , 和 radial basis function kernel (e^[-1/(2*sigma) * (Xi-Xj)
^2] 的 SVM 来成功 separate。
这时候吧数据里面每一个点都向上平移10. 就是 (xi,xj) = (xi,xj+10).
再用以上三个kernel 重新找boundary. 相对于每一个点来说, 新的boundary 会不会
和旧的 boundary 不一样? 也就是说每一个新的点到新的boundary, 和每一个旧点到
旧的boundary的距离是不是不一样。
我觉得linear 的是一样的,应为就是平移。 但是quadratic kernel 和 radial basis
不一样,但是无法具体解释为什么。。。。 |
|
h**********c 发帖数: 4120 | 2 我写这个程序的本意就是能让计算机本科二三年级的学生就理解,并且不需要任何特别
package或算法就可以用c或java把matlab的script重写出来。
这个版在数学以及算法上的高手是很多的,办法肯定有更好的。 由于时间关系, 我没
法试验了。牛顿法以前写过,挺长时间了。
如果说数学上更好的办法,这个点是一个fixed point,能找到一个quadratic
convergence,那么这是最快的。
我的稿还是打算投一投试试,主要希望一些定理,概念能应用到离散的convex hull 中。
大家对我的观点还是比较宽容的,如果您愿意看,过些日子可以发给您。
我打算先试再说,如果您也是相关刊物的editor,我现在就可以给您发。
手上一大堆数据要处理,这个问题需要放一放。
我相信数学方法应该是去构造一个|f''(x)|=0的函数算这个fixed point,quadratic
convergence 估计是不能再快了, 解一个系统用 1e-7左右的精度,通常就三到四个
iteration. |
|
g****t 发帖数: 31659 | 3 我是工程师。很多年没写过文章了。
我写这个程序的本意就是能让计算机本科二三年级的学生就理解,并且不需要任何特别
package或算法就可以用c或java把matlab的script重写出来。
这个版在数学以及算法上的高手是很多的,办法肯定有更好的。 由于时间关系, 我没
法试验了。牛顿法以前写过,挺长时间了。
如果说数学上更好的办法,这个点是一个fixed point,能找到一个quadratic
convergence,那么这是最快的。
我的稿还是打算投一投试试,主要希望一些定理,概念能应用到离散的convex hull 中。
大家对我的观点还是比较宽容的,如果您愿意看,过些日子可以发给您。
我打算先试再说,如果您也是相关刊物的editor,我现在就可以给您发。
手上一大堆数据要处理,这个问题需要放一放。
我相信数学方法应该是去构造一个|f''(x)|=0的函数算这个fixed point,quadratic
convergence 估计是不能再快了, 解一个系统用 1e-7左右的精度,通常就三到四个
iteration. |
|
a*********3 发帖数: 660 | 4 定义 definition变量 variable面积 area直径 diameter半径 radius公式 formula
单价 unit price范围 range/scope/extent集合 set法则 principle本金 principal利
率 interest rate利息 interest单利 simple interest复利 compound interest正数
positive number负数 negative number解析式 analytic expression分类讨论
classified discussion性质 nature (不是很确定)奇函数 odd function偶函数
even function对称 symmetric坐标原点 origin单调性 monotonicity(不是很确定)
任意 random周期性 periodic 有界性 boundedness 数学 mathematics, maths(BrE)
, math(AmE) 公理 axiom 定理 theorem 计算 calculation 运算 operat... 阅读全帖 |
|
g*****h 发帖数: 15 | 5 College Algebra by Larson & Hostetler
DVDRip | MKV / AVC 692 Kbps | 720х544 | 29.97 fps | MP3 128 kbps | 8.75 GB
Language: English
Lecture Titles:
Prerequisites
P-1 Review of Real Numbers and Their Properties
P-2 Exponents and Radicals
P-3 Polynomials and Special Products
P-4 Factoring
P-5 Rational Expressions
P-6 Errors and the Algebra of Calculus
P-7 Graphical Representations of Data
Chapter 1 – Equations and Inequalities
1-1 Graphs of Equations
1-2 Linear Equations in in One Variable
1-3 Mode... 阅读全帖 |
|
k***g 发帖数: 7244 | 6 呵呵,还是数据分布的问题啊,如果都是正值,X 和 X^2 的线性相关总是很高(即便
是你改变scale效果也不明显),除非有负值,这样才能显示出 quadratic relation
来,譬如 -2 -1 0 1 2 和他们平方之间的linear correlation 是 0,这样才能显示
出 linear 和 quadratic 的差异来。 |
|
g******r 发帖数: 29 | 7 well, given 2 random processes X Y
i think we can define the correlation of X Y
as the correl(X_t, Y_t),
if dX_t = a_t dB_t
dY_t = b_t dW_t where B W two brownian motions
the quadratic variation of X_t and Y_t
E[\int_0^t a*b ] = E[\int_0^t a dB * \int_0^t b dW] = cov(X_t, Y_t)
is the quadratic variation of B and W
.
(t |
|
y******i 发帖数: 199 | 8 我的意思是,mean reverting的drift term对quadratic variation有影响,而线性
drift term对quadratic variation没有贡献。 |
|
s*****u 发帖数: 164 | 9 Does anybody have any idea how the FX vol smile
is constructed in Bloomberg? I tried to use 25
delta market quotes to fit with quadratic delta
vol, vanna-volga, and SABR. Quadratic delta vol
gives lower volatilities for options deep out of
money, and VV and SABR gives higher results. I
also tried to find any documentation for the recipe
used in Bloomberg, but could not find anything.
Can anybody help me on this matter? Thanks! |
|
z***y 发帖数: 1 | 10 搭车同请教,解quadratic programming with quadratic constraints有什么package
比较好吗?多谢! |
|
o******6 发帖数: 538 | 11 ☆─────────────────────────────────────☆
orioniori (猎人) 于 (Sun Mar 2 16:20:58 2008) 提到:
发信人: orioniori (猎人), 信区: EE
标 题: Nonlinear Constrained Optimization 问题求教
发信站: BBS 未名空间站 (Sun Mar 2 14:18:31 2008)
在下是新手,不才, 有初级问题请教达人们:
目标函数 F(x1,x2,....xn)可以closeform表示出来,是个非线性函数(很多erfc函数
和指数运算,非Quadratic)。
约束函数T(x1,x2,....xn)也是一个非线性非quadratic的函数,并且可以closeform表
示出来。
现在给定Tmax上限常量,要求:
[x1,x2,....xn] = argmax{F(x1,x2,....xn)}
s.t.
T(x1,x2,....xn) <= Tmax
有什么比较好的方法,除了穷举之外?
有什么比较好 的reference或者tutorial可以参考? |
|
p******k 发帖数: 23 | 12 logistic loss不是quadratic, 也不是piecewise linear. 根据 prof. Ji Zhu的文章
,它的solution path不是piecewise linear的,应该是非线性的。 原则上讲和Least
Angel Regression 是不一样. 但是我们总是可以用local quadratic去逼近 logistic.
所以还是可以用small step forward很好逼进的 |
|
l*****y 发帖数: 5 | 13 I have such type of matrix problems: An unknow matrix X \in R^{nxn} satisfies
an equation with X in quadratic form (power to two). For example, X is real
symmetric and the equation is:
( P * X * P' + Q ) = X * R * (P * X * P' + Q ) + X
P, Q, and R are given constant matrices. My question is: what is the approach
to generally solve such type of equations? it is really a simple quadratic
equation for any scalar X (middle school problem). But how about if X is an
unknown matrix? Is such type of mat |
|
t******l 发帖数: 10908 | 14 当然我说的意思,不是说任何能够对优化有帮助的千老 paper 都可以。。。理论的作
用一般得有一个维度的时间复杂性进步,或者至少也得是半个维度(比如 quadratic
到 super-linear 啥的)的进步。。。否则的话,还是千老要吃饭型。。。因为小于
半个维度的话,如果运气好,我啥都不用干,搬真皮大沙发坐等 INTEL 下一代处理器
出炉就行。。。运气差一点的话,搞他妈几个工业界 trick 搞一搞竞争对手就可以了
。。。 |
|
t******l 发帖数: 10908 | 15 从 quadratic 到 superlinear,在工业界大尺度运算上,是革命性的。
: 算法复杂度从n^2到nlogn。发paper不算啥。具体到工程上enable了无数可能。
“我们
: 一生中最重要的数值算法”,20世纪十大算法。1805年由高斯提出。
: 我最早在图像处理上遇到快速傅里叶变换。
|
|
b*****d 发帖数: 61690 | 16 北京时间9月7日,中国大陆媒体华尔街见闻援引道琼斯的消息称,海航据称退出在德银
的持股。 公开报道称,海航原为德银最大股东,此前由于资金缺口的问题已陆续透过
海航旗下奥地利资产管理公司C-Quadrat减持德意志银行至8.8%。 |
|
a**e 发帖数: 8800 | 17 [Benjamin Solomon "Ben" Carson, Sr. (September 18, 1951)] Source: LYBIO.net
Thank you so much. Mr. President, Mr. Vice President, Mrs. Obama,
distinguished guests – which included everybody. Thank you so much for this
wonderful honor to be at this stage again. I was here 16 years ago, and the
fact that they invited me back means that I didn’t offend too many people,
so that was great. [LAUGHTER]
I want to start by reading four texts which will put into I want to start by
reading four texts which... 阅读全帖 |
|
a***s 发帖数: 5417 | 18 Here's the transcript of his speech on national breakfast.
-----------------------------------------------------------
[Benjamin Solomon "Ben" Carson, Sr. (September 18, 1951)] Source: LYBIO.net
Thank you so much. Mr. President, Mr. Vice President, Mrs. Obama,
distinguished guests – which included everybody. Thank you so much for this
wonderful honor to be at this stage again. I was here 16 years ago, and the
fact that they invited me back means that I didn’t offend too many people,
so that was ... 阅读全帖 |
|
t*******t 发帖数: 1067 | 19 25 ft jump shot vs solving quadratic equation.这就是当面打O8的脸, |
|
h**********g 发帖数: 3962 | 20 一下是图灵官网的描述。我觉得他在密码学方面的贡献是很重要的一部分。
Yao spent a year as a Professor in the Computer Science Division of the
University of California, Berkeley, and subsequently returned to Stanford as
a full Professor in 1982. During the early 1980’s, Yao produced a number
of papers which had a lasting impact on the foundations of cryptography,
computer security, computational complexity and randomized computation. This
work was significant not only for results obtained, but also for the
introduction of problems,... 阅读全帖 |
|
s*******s 发帖数: 1568 | 21 use newton's method to achieve quadratic converge.
of |
|
l*****p 发帖数: 11 | 22 Had an interview with them 2 years ago, asked to implement Hashtable with
quadratic probing. I didn't do too well.... |
|
k***e 发帖数: 556 | 23 kao 好变态
quadratic probing如果超出数组界限是继续到数组头再开始吗?
一下子还真想不起来了 |
|
m******9 发帖数: 968 | 24 hash function通常有:
modulo
folding
mid-square function
extraction
radix transformation
collision的解决办法:
line probing
quadratic search
double hashing
chaining
bucket addressing |
|
S******n 发帖数: 1009 | 25 yep, newton's method is perfect for quadratic problem
能保证任意初值 |
|
M7 发帖数: 219 | 26 面试里也就粗略地说一下,linear, NlogN, quadratic ....
那些NP, P-space ... Polynomial hierarchy的东西只有做理论的人才研究的。
不过回想当时钻研Garey & Johnson的时候,还是挺好玩的。 |
|
y***o 发帖数: 88 | 27 【 以下文字转载自 Working 讨论区 】
发信人: yunho (likeyostyle), 信区: Working
标 题: job opening in Phoenix, AZ (Apollo Group Inc)
发信站: BBS 未名空间站 (Tue Aug 16 18:34:50 2011, 美东)
Position Title: Operations Research Scientist
Responsibilities:
• With an understanding and knowledge of applicable business
systems and industry requirements, in conjunction with proficient
knowledge
in Operations Research techniques, this position is responsible for
developing and devising advanced decision technologies, analyt... 阅读全帖 |
|
|
|
d********t 发帖数: 9628 | 30 不是,似乎是做了一个Minimizaion得出的结果。 |
|
l******c 发帖数: 2555 | 31 Why computers take binary input? |
|
|
n*******w 发帖数: 687 | 33 有点印象。好像有证明,算出来是2点几。2或者3比较好。2更简单。所以2了。 |
|
|
n*******w 发帖数: 687 | 35 可能是吧。数字不记得。但是2比3简单,所以2进制,2-tree了。 |
|
b****g 发帖数: 625 | 36 triple/quad/any other trees can be represented in binary tree
But I am not sure this is the correct answer |
|
|
b******t 发帖数: 965 | 38 onsite google的时候也碰到个 问sqrt的 给他讲了半天 newton法 什么是quadratic
convergence
为什么binary search 是linear convergence (因为给的input相当于是logN个bit)
那个人似乎完全没见过这个 然后还是要我写binary search法 |
|
p*****2 发帖数: 21240 | 39
quadratic
你整个一个over qualified. 我要是面试官,你会吓死我的 |
|
g*********e 发帖数: 14401 | 40
quadratic
废话 你写个他不懂的 就算写错了也查不出来。他不能任你摆布 |
|
j******2 发帖数: 362 | 41 多处看到这个结论。但是想不通啊,难道不是要quadratic time吗?比如所有字符都不
一样。 |
|
j******2 发帖数: 362 | 42 每检测一个元素时不是同时也看了它是不是到队尾了吗?所以不会扫超过min(m)轮。
另外我在想:divide and conquer是不是只能把quadratic变成log linear,对本身
linear的问题没有帮助啊? |
|
G****A 发帖数: 4160 | 43 都是大家熟悉的题
Q1, nth Fibonacci number. 要求尽可能多的解法。
follow-up, 每种解法的time/space complexity (要求function call的stack
overhead也看做space complexity).
Q2: c++基本概念。static, reference vs pointer, what is OOD, ...
Q3:列举并比较你常用的几种design patterns的作用和特点。
Q4: tree traversal, 要求level by level,按从底向上的顺序。同个level从左到右。
当我说要用DFS实现print by level的时候,此senior engineer先说“How could you
possibly print in level-order using DFS?”
解释完之后,senior engineer说:“ In the worst case, this will have
quadratic run time (if the tree is unbalanced)... 阅读全帖 |
|
A******i 发帖数: 34 | 44 【 以下文字转载自 EE 讨论区 】
发信人: AngelaQi (Angela), 信区: EE
标 题: Electric Engineer-Santa Barbara- salary: 80,000-100,000/ye
关键字: Electric Engineer,Santa Barbara
发信站: BBS 未名空间站 (Thu Sep 5 19:39:32 2013, 美东)
Seeking Senior Electrical Engineer who excels in Motor control, Analog /
Digital Design, and Board Layout. Applicant will be part of an elite team
capable of developing a variety of leading edge semiconductor
instrumentation.
SUMMARY: Responsible for the electrical aspects of design, development and
m... 阅读全帖 |
|
|
n****o 发帖数: 239 | 46 首先你要把数据plot一下看看数据怎么样子,然后一步一步来:有没有周期,有没有趋
势,什么趋势(linear, quadratic, etc.)等等。把这些考虑进去后,要清除这些,剩
下平稳时间序列,用现有的包来fit.预测的时候把先前消除的那些周期,趋势加上来,
就可以预测了。Good luck. |
|
f*******s 发帖数: 182 | 47 O(1) = constant time/space, oh-one
O(logn) = logrithmic time/space, log-n
O(n) = linear time/sapce, oh-n
O(nlogn) = n-log-n
O(n2) = quadratic time/space, n-squared
O(n3) = n-cubed
O(2n) = exponential, two-to-the-n
O(n!) = n factorial |
|
w*******g 发帖数: 9932 | 48 just printing the results takes exponential time in worst case. how can it
be quadratic time.
I agree that dynamic programming is the approach
find results for size 2,4,6,... is sufficient |
|
w*******g 发帖数: 9932 | 49 just printing the results takes exponential time in worst case. how can it
be quadratic time.
I agree that dynamic programming is the approach
find results for size 2,4,6,... is sufficient |
|
发帖数: 1 | 50 啊,3大概是F或者ID PF or whatever,因为可能有competing risk,可能有quadratic
trend。 |
|