发帖数: 1 | 1 真心恐怖！！！
https://www.zhihu.com/question/59407631
作者：苏闵
链接：https://www.zhihu.com/question/59407631/answer/168650246
来源：知乎
著作权归作者所有。商业转载请联系作者获得授权，非商业转载请注明出处。
个人觉得最理想的状态是跟一个处于快速上升期的导师，也就是有潜力的新人。讲个我
自己的经历。我在英国读博，博一读了小半年从一个大牛的组转出来了，转到一个中等
的组，现老板是个起步级/中级的正教授，且方向相对冷门。除了原组内的人没有人理
解我，但我不后悔。原组的导师是锂电池方面世界前十的大牛（加上他是英国人这个信
息业内人应该知道他是谁）。我去面试的时候经历就有之后工作节奏的前兆——从原定
的时间生生等了一个下午，直到快晚饭时间他才出现。等待的过程中组里的学生和博后
都有意无意地跟我说组内的情况很复杂，让我最好考虑别的选择。后来整个面试过程就
10分钟，他表现得很nice，问了几个很基础的专业问题就过了。我没抵御住导师名声的
诱惑（还有一些个人原因），最后还是接了offer。入学之后发... 阅读全帖 |
|
W******r
发帖数: 789 | 2 来自主题: ChineseMed版 - 新天方夜谭 CDC Panel Votes to Extend HPV Vaccine to Young Boys
All males starting at age 11 should receive the HPV vaccine Gardasil to
protect themselves against sexually transmitted forms of human
papillomavirus, the cause of most cervical and anal cancers as well as most
mouth and throat cancers, a Centers for Disease Control and Prevention
advisory committee voted today.
Thirteen members of the committee voted in favor of extending the HPV
vaccine recommendation to young boys, and one member abstained. ... 阅读全帖 |
|
j********n
发帖数: 792 | 3 A survey of the foundation was performed. The overall tilt of the foundation
was less than 0.15% as evaluated along the major axis. Several American
Society of Civil Engineers (ASCE) Foundation Subgroups consider tilt of less
than 1% acceptable. The local floor slope of the foundation was less than 0
.1 % in the front bedroom and other areas of the residence. Local floor
slope of up to 0.5% is acceptable to FHA and other lenders. The interior
distress observed within the residence was cosmetic a... 阅读全帖 |
|
f******r
发帖数: 2975 | 4 Lattice is a discrete subgroup of Euclidean Space
Suppose I have one lattice L now,
choose some deep holes of the lattice P0,P1,..Pn
Construct L'= L+a_0P0+a_1P1+...+a_nPn
where a_0,a_1,...a_n are either 0 or 1.
Now I am thinking whether L' is a lattice
I guess it is but I don't know how to prove it
Here is my step for addition
Let L1,L2 be two points of the lattice L
L1'=L1+a_0P0+a_1P1+...+a_nPn
L2'=L2+b_0P0+b_1P1+...+b_nPn
L1'+L2' = L1+a_0P0+a_1P1+...+a_nPn+L2+b_0P0+b_1P1+...+b_nPn
then I don't |
|
T*******x
发帖数: 8565 | 5 最近刚接触抽象代数，问题不少。
我在想这个问题：实数加法群，有理数加法群，整数加法群
(R,+)，(Q,+)，(Z,+)之间的同态关系：
1,(Q,+)-->(Z,+)只有平凡同态，即所有有理数都映射到0
这个我已经证明了。
2,(R,+)-->(Z,+)应该只有平凡同态。
3,(R,+)-->(Q,+)应该只有平凡同态。
2,3,我还没有证明出来。它们又引出一个问题：
4,(R,+)有没有不可数真子群(uncountable proper subgroup)？
如果没有的话，那么2,3就成立了。 |
|
b**g
发帖数: 335 | 6 拜托你去看一下书里关于finite field的multiplicative subgroup是cyclic的证明
根本不是你那样证的,根本也没用到费马小定理,没你说的那么trivial
e.g Dummit & Foote p.315 Corollary 18,19 |
|
z**d
发帖数: 11 | 7 Let G_1, G_2 be two subgroups of G. X_{i} be an arbitrary partition of G.
G=\cup X_{i}
clearly, there exist some i, such that $G_1 \cup G_2 \in X_i$.
For any $g\in G_2$, we have $g\in gG_1$. But $g\in X_i$, it follows that gG_
1\in X_i.
Therefore
\in X_i
\qed |
|
W**********d
发帖数: 15 | 8 Well, I guess Z_2 is a simple group. Suppose your group has a subgroup with
order m, it is not simple. |
|
a*******o
发帖数: 2 | 9 Let G be your group with order 2m, where m is odd.
It is well known that any even-ordered finite group contains
an element of order 2. Let H be the set of elements x in G
satisfying x^2 = e (the identity element). Then the order of
H is an even number and a divisor of 2m. Thus, |H| = 2 or |H| = 2m.
If |H| = 2m, then H = G. One can then show that G is abelian
and, with order =2m, not simple.
If |H| = 2, then one can show that H is a nontrivial
normal subgroup of G. Thus G is not simple.
Some deta |
|
g******a
发帖数: 69 | 10 by considering the pair (a,a^{-1}),
we can find g_1\in G such that g_1\neq e and g_1^2=1.
Next, let A be the commutation group of G, i.e. |A|=(2m)!.
For any g\in G, map the action on G of left multiplication
by g to A: g\to T_g\in A, T_g(h)=gh.
T(g_1) is a commutation of odd order.
Let G_1 be a subgroup of G such that T(g) is of even order for
any g\in G_1. It is easy to check that G_1 is a normal group
and |G_1|=m, since if T(g) is of even order, then T(gg_1) is
of odd order. |
|
f******h
发帖数: 104 | 11 来自主题: Mathematics版 - 代数群问题 Linear Algebraic Groups
James E. Humphreys
an exercise on p.57
A closed subset of a linear algebraic group
which contains e (the identity element) and
is closed under taking products is a subgroup. |
|
l******e
发帖数: 4 | 12 来自主题: Mathematics版 - 代数群问题 Say the group is G, and the closed subset is X. Consider the right regular
representation of G in K[G]. Take a finite dimensional G-invariant subspace
V of K[G] such that W=V\cap I(X) generates I(X). Let H be the stabilizer of
W in G. Then H is a subgroup. It is not difficult to show that H=X. |
|
f******h
发帖数: 104 | 13 Prove that PGL(n) is isomorphic to a closed subgroup of GL(n^2),
where PGL(n) is the projective linear group.
Thanks. |
|
l*****e
发帖数: 65 | 14
^2
I will try to give an answer, but really not sure about its correctness. So
use your own judgement to find out the truth.
My answer is: any number n. (b=a^{-1} for n=1, and we discuss n>1.)
The reason: build up an example of a solvable group for the number.
Given any n>=2, let H= the order n cyclic group. Consider an
involution y acting on H by y^{-1} x y = x^{-1}.
Let G be the group generated by semi-product of H by y. Namely,
G=
Then 1. G is so... 阅读全帖 |
|
L*m
发帖数: 235 | 15 最近十年在annals of mathematics上发表或合作发表文章的华人全统计（不包括
terrence tao和一位mit本科毕业的abc华人），单位统计以现在作者单位为准
annals
2015年
A proof of Demailly’s strong openness conjecture
关启安（北京大学） 周向宇（中科院）
A solution of an L2 extension problem with an optimal estimate and
applications
关启安（北京大学） 周向宇（中科院）
Finsler metrics and Kobayashi hyperbolicity of the moduli spaces of
canonically polarized manifolds
杨世琪（普渡大学） Wing-Keung To（新加坡国立大学）
Construction of Cauchy data of vacuum Einstein field equations evolving to
black holes
黎俊彬（中山大学）... 阅读全帖 |
|
m***0
发帖数: 39 | 16 Great idea! One suggestion: could you have different subgroups for
different stages: beginners, mid levels and advanced levels. |
|
s*h
发帖数: 1538 | 17 here are some questions I have got so far:
1.why this program (all asked this)
2.tell me about yourself (whatever i prepared before did not work for me...I
kept changing my answers in each program;hopefully they wont catch the
discrepancies in my answers, lol)
3.why this specialty
4.how do you deal with adversity
5.how do you deal with a failure
6.how did you manage your personal life with the pressure of the medical
school(I did not obviously but , again they dont know that!)-I just
graduated..... 阅读全帖 |
|
B****a
发帖数: 1526 | 18 http://www.nature.com/news/2011/090211/full/470141a.html
Pharmaceutical industry must take its medicine
To fix the drug pipeline, governments must take on drug-makers instead of
capitulating to their every demand, says Colin Macilwain.
Pfizer's announcement last week that it is to pull the plug on its drug-
development laboratory in Sandwich, Kent, and fire most of its 2,400 staff (
see page 154), must be a wake-up call for scientists and policy-makers alike
. The pharmaceutical industry is taki... 阅读全帖 |
|
s*********x
发帖数: 67 | 19 同意LZ的看法,FDA和EMA对于risk benefit profile的评审已经达到非常细致的程度了,
对不同TA的药物要求甚至会相差很大,所以不能和原文作者描述的那样一概而论.
原文作者把临床实验的设计想得过于简单:
1.首先是TA不同要求不同:
比如oncology对于I\II\III的要求,某些方面就比较宽松,可以使用一些surrogate
endpoint,临床样本也相对比较小(当然这个也和统计要求的power和non-inferiority
margin有关)
但同时如果你看cardiovascular, 患者样本量很多都是成千上万,对于risk/benefit
asseement的要求也不一样,对于primary endpoint要求也常见全因死亡这种大终点,很
少有surrogate
2."临床试验是在一定的人群中进行，而不同的人群对实验化合物或者生物制品（比如
说疫
苗）的反映可能不同，所以即使是在整个人群中效果不好，而在某些人群中效果好，但
FDA不会因在某些人群中效果好就批准这个实验化合物或者生物制品，就是要根据原来
设计时的人群来看结果"
LA所说的"某些人群... 阅读全帖 |
|
l****o
发帖数: 32 | 20 Please apply for it online ASAP if you are interested in this opportunity
and I will recieve your resume right after that. Thanks.
Auto req 30057BR
Job Posting Title Grad Intern - Operations
Career Category College Job
Employee Subgroup Seasonal FT
Country (State/Region) United States - California
Location (City) US - CA - Thousand Oaks
Amgen Job Description Amgen’s 10 – 12 week internship program offers
meaningful project experiences that impact patient’s lives. This... 阅读全帖 |
|
l******k
发帖数: 27533 | 21 Statin的作用太多RTC证明了。
Fish oil，在GISSI trial 里的subgroup analysis有证明positive effects for
cardiovascular disease and mortality. 在2013 ACCH/ACA CHF guideline里有推荐
fish oil 给stage C HFrEF, 但是是class II B evidence,实践中我们不推荐！
Red rice yeast 当然不如处方statin好，但好歹也是个statin。 OTC的fish oil 要一
天吃多少颗才能达到有效计量你想过没？
★ 发自iPhone App: ChineseWeb 8.6 |
|
j*********t
发帖数: 217 | 22 不是G8，是E8
Jacques Distler有一篇文章批驳这个
http://arxiv.org/abs/0905.2658
There is no "Theory of Everything" inside E8
Jacques Distler, Skip Garibaldi
(Submitted on 16 May 2009)
We analyze certain subgroups of real and complex forms of the Lie group E8,
and deduce that any "Theory of Everything" obtained by embedding the gauge
groups of gravity and the Standard Model into a real or complex form of E8
lacks certain representation-theoretic properties required by physical
reality. The arguments themselves |
|
m*t
发帖数: 2464 | 23 1-17
一
暑假看西游记，发现其实讲的是一个年轻的科研人员的成长经历。
唐僧
如来院长的学生（二徒弟金蝉子），博士生期间工作平平，只发了几篇弱的sci，混了
个毕业。想要留在西天学院比较困难。
如来院长就对观音院士说：你看啊，最近有“西天取经”这么个项目，很有前途，你可
以找一个有能力的年轻人来搞。
观音院士会意，就说：唐僧这个人不错，吃苦耐劳，科研能力出众，我收为posdoc吧，
给他组个subgroup。
其实观音院士是明白的，唐僧这个人虽然师出名门，但是忒老实，实在没有那种开拓的
闯劲，于是给他物色了几个比较猛的学生。
孙悟空
大牛菩提老祖老板的学生，但是因为太急功近利，哪了个七十二变搞了篇pra就硕士毕
业了。菩提老板知道这个学生不安份，在他毕业的时候就说：你以后不要给人说是我的
学生。
孙悟空则自以为很拽，回到家乡的民办高校花果山学院当了教授。跟牛魔王蛟龙王一群
民科混的很high。到了后来还经常comment水族大学东海学院院长敖广教授和地府大学
校长阎王教授的文章，搞得敖老板阎老板很不爽，拉着他们的老师天庭研究所所长玉帝
院士要跟孙悟空论战。
太上老君院士比较有见识，说... 阅读全帖 |
|
h*****t
发帖数: 1478 | 24 Cleckley当时的说法并没有区分psychopath subgroup.
其中一部分ANTISOCIAL的人智商可能不高，甚至偏低。
相反，在各种领域都很successful且没有反社会表现的人相对来说智商不会低。而且这
里说的智商较高，是跟一般人比，120－130已经很高了。
intelligence
AP |
|
b***k
发帖数: 2673 | 25 ☆─────────────────────────────────────☆
heavenapple (heavenapple) 于 (Mon Dec 10 22:43:01 2007) 提到:
发信人: heavenapple (heavenapple), 信区: Statistics
标 题: 今天一道面试题目。大家帮忙看看。
发信站: BBS 未名空间站 (Mon Dec 10 22:40:16 2007)
比较advantage/disadvantage of methods of finding standard error.
这个是什么意思呀？
难道还有很多方法求se的方法么？不就是按定义来求么？
☆─────────────────────────────────────☆
hjdut (hehe) 于 (Tue Dec 11 16:10:23 2007) 提到:
定义是一种，还有好像先subgroup，不需要random error normal的条件似乎，你查一
查paper,肯定有
☆──────────────────────────── |
|
a***r
发帖数: 594 | 26 GS Strat group is an independent group. It has a few hundred strats, most of
whom are distributed to subgroups working with trading desks, sales, IB,
asset management.
In the core culture of strat group is the SLANG scripting language. It is
said to be the glue that keeps strats together even though they work across
a wide spectrum of businesses.
Core strats are charged to develope and maintain SLANG. And they are also in
charge of developing tools used by strats in multiple businesses, like a
g |
|
s*********y
发帖数: 45 | 27 谢谢大家的回复。 看了回帖之后补充一下， 我觉得Ballack5说得很对，我去了之后基
本上是在 infra 和 trading team 之间，过段时间可能会转 trading. 有trading
team 跟我电话聊过，但我觉得没有面谈了解不够不想直接jump in. 我去的组是
analytic group 里面一个subgroup, 现在七个人，说是做 "research", 具体弄什么我
也不是非常清楚，貌似是研发一些strategy包装成软件，然后看看有没有trading team
想用。Tower的package实在诱人，除了工资高还有 6 weeks paid vacations, 但是去
HFT有其自身风险，而且exit options 估计比不上GS, 所以纠结。 |
|
f*******3
发帖数: 577 | 28 你确定analytic group里面还分subgroup？据我的了解，貌似只有一个analytic group
吧。。。
team |
|
s*********y
发帖数: 45 | 29 这个倒真没。
一开始两个offer的时候是准备去高盛的。但是和HR谈到具体package的时候发现Tower
的package好太多了，所以现在难决定。去GS之后title只是 PhD associate, 具体是什
么position其实我也不清楚，合同也没明写。只知道是会在两个组里rotate一段时间，
然后自己决定。和两个组的头子都聊过挺久，但是他们就光讲很多他们做的都是些啥事
，具体我做什么他们没细说我也没细问，反正开始也就是到处打杂和学东西。肯定不是
Trader, developer，business analyst. 不知道是quant 还是 desk strategist， 能
说说区别么？ 见笑了，我真是新人，很多都不懂。 Tower的结构相对Goldman简单，所
以知道自己要去那个组内的subgroup，Goldman就太复杂，前前后后谈了十多人我根本
搞不清楚也记不得谁是哪个组的，只记得两个头子是谁。 |
|
D**u
发帖数: 204 | 30 弄了半天搞出个证明来, 自己看着都不觉得漂亮, 就将就着读吧.
To make it clear, I think you mean: If the group is 2n people,
then everyone know >=n people; if 2n+1 people, then everyone
know >=n+1 people (not include himself).
Using induction:
1) The problem is true for 3,4 (easy to discuss)
For 2, if 2 people know each other, we still think it to be
a circle.
2) Suppose the problem is true for people up to 2n.
3) Let's think about the case when group has 2n+1 people.
People are A_1, ...A_(2n+1). In the subgroup A_1,.. |
|
D**u
发帖数: 204 | 31 Suppose the group is G, define a function f(A) on all the subgroups
A of G as: f(A) = the # of relationships between A and G-A.
Notice that f(A)=f(G-A)
First you arbitrarily seperate G into 2 groups A and G-A.
If one guy has more than half(strictly bigger) friends
in his own group (suppose this guy is "a", and in group A)
Then f(A-{a})>f(A)
Let A1=A-{a}.
Keep doing this, you will get f(A)
since f takes integer values, so you will stop the process at
one point, the you get the 2 subgrou |
|
y***u
发帖数: 25 | 32 Unitary group is SU(N), but Lorentz group is SO(3,1). You can check if Lorentz
group is unitary or not by a particular representation. Say, in the spinor
representation, the generator is {\Sigma}^{u v}=i/2
{\gamma^{u},{\gamma^{v}}+x^u P^v-x^v P^v. We know \gamma^u{\dagger}=\gamma_u
(the metric is +1,-1,-1,-1). By using the anticommutation of gamma matrices,
you can see as u,v=i,j (spatial component) the generators are hermitian which
means the corresponding subgroup is unitary. But as u,v=0,i, t |
|
q**j
发帖数: 10612 | 33 nothing surprising. they should not be the same. first group then taking ave
rage is not the same as taking subgroup average, group and then average agai
n. |
|
t*********e
发帖数: 313 | 34 By sample size, i mean the total count by subgroups in a two-way table.
Thanks! |
|
b*******r
发帖数: 152 | 35 food for thought--
1. grouping/subgrouping/segmentation is a good answer. another direction is
to do the transform of x1, eg. log(x1).
2. i think he is looking for some variable reduction techniques, like
clustering, pca, fa, etc...
3.4.i think they are more likely open questions.
5.can u say say why is yr answer?
thx |
|
l*******n
发帖数: 19 | 36 "Random sample" and "Randomized design" are different concepts.
If there is no treatment intervention, you have an observational study.
Your samples (no matter old or not) are still random.
If there is treatment intervention, the subgroup design (old people) may
not be randomized even your overall design is randomized, when your design
was not stratified by age (with old as a stratum). |
|
D******n
发帖数: 2836 | 37 A I L R
a 0 1 2
a 0 3 4
a 1 5 6
a 1 2 3
a 0 4 5
a 0 5 9
b .....
b .....
b ....
...
||
\/
A I L R
a 0 1 4
a 1 5 6
a 1 2 3
a 0 4 9
b .....
b .....
b ....
...
basically within each A group(such as a,b,c,...), there are 0s and 1s for
the I attribute, i want to merge the each consecutive 0 subgroup(for example
the first 2 obs) into one entry keeping the L from first obs and R from
last obs. |
|
b2
发帖数: 427 | 38 proc sgplot data=
var region / group=size response=total stat=sum;
run;
response 只允许跟一个变量，total=coount before 2000+count after 2000;
如何将coount before 2000 和 count after 2000加进去，从而比较三者。
我曾考虑过proc gchart,也不行啊，我的是三列变量，不是subgroup啊。
除非我把他们分类求和，
想问问有没有简单的方法，谢谢~ |
|
c**********5
发帖数: 653 | 39 HI,你们好，
这个是我的临床实验设计。
医院----randomization-----A or B---在每个subgroup的病人，再进行
randomization，C OR D。在接下来的12 个月内，每个病人有3 次repeated
measurements for the outcome response。
Q1。 想知道intervention effect on outcome.
OUTCOME1(BINARY)
proc glimmix ;
class Hosp_id time trt_grp patient_id;
model outcome=trt_grp|time/link=logit dist=binomial solution;
random intercept/subject=Hosp_id type=un solution;
random time/subject=patient_id type=cs residual S;
run;
Fist it i... 阅读全帖 |
|
s*r
发帖数: 2757 | 40 you are exploring into an area called multiple testing. these p-values from
subgroup analysis can be totally meaningless |
|
l*******s
发帖数: 437 | 41 但是这里los并不是整数，subgroup size也不同。我们已经用了run chart，他们想看
trend。又一定要我用upperlimit和lowlimit要有变化的那种，我用的XRCHART，他们一
看就齐刷刷的两条线就不高兴了。
还有其他好的选择吗？ |
|
D*D
发帖数: 236 | 42 想要把一个sorted dataset里，每个subgroup的前三observation输出。
test code 如下
data temp;
do i = 1 to 5;
do j = 1 to 5;
output;
end;
end;
run;
proc sort data = temp; by i j; run;
data new;
set temp;
by i;
if first.i then do;
do k = 1 to 3;
set temp;
by i;
output;
end;
end;
run;
proc print data = new;
run;
结果是
Obs i j k
1 1 1 1
... 阅读全帖 |
|
j****n
发帖数: 10 | 43 To your old code:
15 records is the right output.
the 1st 'set temp;' Decide the logic true or not (used in the if condisiton)
the 2nd 'set temp;' if the conditin is true, then 3 records will be read.
for the 1st time, the condition is true, then record 1 2 3 will be read.
for the 2nd time, the condition is true, then record 4 5 6 will be read. (
the position of reading the record is not 6, which is the first of subgroup
2).
If you change you code 'first.i' to 'last.i', you will still get 15 rec... 阅读全帖 |
|
D*D
发帖数: 236 | 44 Thanks for the explanation here. It really helps.
The code you modified was not right though. It printed the same combination
of i and j for 3 times.
What I was looking for was to output the first 3 obs of j, that is j = 1, 2
and 3.
I came to the following code which does the job. Thanks for the discussion.
I have a better understanding of data step now.
data temp;
do i = 1 to 5;
do j = 1 to 5;
output;
end;
end;
run;
proc sort data = temp; by i j; run;
data new (drop = t);
retain t;
set... 阅读全帖 |
|
d******9
发帖数: 404 | 45 Try to use group and subgroup options. |
|
x*******u
发帖数: 500 | 46 proc gchart 横坐标是date by month, 数据很多, 画出来, 坐标下面的字密密麻麻看
不清楚. 请问, 怎么只显示年份.
其他proc 也行. 反正就是画个bar char, 横坐标是date, 纵坐标是count.
我试过
axis1 order=("01Oct98"d to "01Dec11"d by year);
proc gbarline data=Plotdata;
vbar date/subgroup=type discrete sumvar=count
maxis=axis1;
format date DATE7.;
run;
quit;
但是bar也随之减少. |
|
s******0
发帖数: 1269 | 47 第一个,没明白你的design structure.你想说的是有random*random interaction?
第二个,你觉得这2段code想要做的东西一样吗?不一样的话结果为什么要一样呢? |
|
g******i
发帖数: 118 | 48 楼主的问题是不是这样？
楼主是有一个panel的数据，数据里面有比如sales，site visit (0/1), 然后vistor的
demographics比如gender等等。目标是：用这个Panel的数据，来估计整个population
的visit和sales的数据。但是，因为panel里面的sample有bias, 我们要根据panel与
population在demographics上面的差异，来adjust estimates的bias，对不？
如果是上面这个问题，我觉得就是经典的sampling and survey的问题，不过很多学校
不专门开课教。其实就是根据demographics, 给panel里面不同的subgroup assign不同的
weight，最后aggregate起来estimates就是unbiased. Weight是调整bias的关键。
想不清楚为什么需要，或者具体怎么用复杂的model比如classification等来解决这个
问题。 |
|
c***z
发帖数: 6348 | 49 【 以下文字转载自 DataSciences 讨论区 】
发信人: chaoz (面朝大海，吃碗凉皮), 信区: DataSciences
标 题: [Data Science Project Case] Bias Correction - third try
发信站: BBS 未名空间站 (Tue Feb 11 18:26:40 2014, 美东)
Dear all, thank you so much for your earlier inputs! Now I am able to put my
thoughts together and understand the project better.
Let me write down the thing again. Any comments are extremely welcome!
Project name: Bias correction
Business objective: We have a panel of 25M users’ shopping cart information
, we want to in... 阅读全帖 |
|
t*****a
发帖数: 459 | 50 嗯 如果只是safety，感觉确实没啥意思。如果能做subgroup analysis，causal
inference, cost-effectiveness, 等等，就会感觉有意思多了。
本来觉得大公司好处之一就是统计师不用自己编程啊，唉。不过我一个mentor说，大公
司还是有一定好处的，起码以后放在简历上，视觉冲击力就比小公司强多了。不过我个
人还是有点担心大公司给硕士的发展空间可能少些，小公司风险和机会都多些。 |
|
