K****D
发帖数: 30533 | 1 Your point of view does not represent the consensus of the majority
of financial advisors. In fact, deciding or projected tax bracket
(as mentioned earlier in this thread) is indeed a prevailing idea
(especially among American advisors).
The common conclusion is 28% and above marginal tax rate should
definitely go with traditional, while 25% should still consider
going traditional. 15% though, should go with Roth.
Typically life expenses after retirement is significantly lower
than those before ... 阅读全帖 |
|
k***n
发帖数: 3158 | 2 I went all in gold/silver last year on Feb. 08 last year when gold was 1080
and silver was 15-, and has been adding.
Finally convinced my parents to buy a big chunk last summer before the
big rally.
could not be happier for this "investment" |
|
k***n
发帖数: 3158 | 3 【 以下文字转载自 Collectibles 讨论区 】
发信人: kafen (Tiffany & Co), 信区: Collectibles
标 题: Re: 白银太惨了，11分钟跌10 percent (转载)
发信站: BBS 未名空间站 (Mon May 2 13:04:32 2011, 美东)
he sold his silver in 2006
at the same time, SLV started and held a similar amount of silver he sold
coincidence?
SLV's trustee is BNY Mellon
guess who owns a big chunk of BNY Mellon?
发信人: xt (拷贝猫), 信区: Military
标 题: Re: 白银太惨了，11分钟跌10 percent
发信站: BBS 未名空间站 (Mon May 2 09:11:48 2011, 美东)
2006年前后他说在$4的时候进了不少，但是到了$6就卖掉了 |
|
a*******d
发帖数: 4846 | 4 Be careful HOA fee eats a big chunk of your profit. |
|
a*******d
发帖数: 4846 | 5 HOA fee stands for "home owner association fee", it will be the biggest
chunk of mandatory fee a condo owner has to pay every month if you have no
mortgage.
It spells as HOA not "HAO". |
|
f*******n
发帖数: 12623 | 6 Your situation about the contribution limit applies only when you actually contribute up to the limit. If you compare amounts that you can fit into either Roth or traditional, e.g. $7.5K Roth vs. $10K traditional, then there is no difference between the two in your method.
One thing you're forgetting is how the tax bracket system works. Remember
that even when your tax bracket is 25%, your initial income is always taxed
at 10%, and then another chunk is taxed at 15%, and then 25%, etc. When you
... 阅读全帖 |
|
k***n
发帖数: 3158 | 7 but they are the majority
a fly meat from a majority can still be good chunk for some "advisors"
especially those who many not land many high-net-worth individuals as
their customers. lol
> |
|
r****m
发帖数: 1204 | 8
pandabig -
Thanks for showing the return numbers.
However, I want to point out this is a simple, but not the right way to
compare IUL and Term+Index, because each year, IUL got chipped away a large
chunk of expenses and cost of insurances. Even at the same % returns of the
two over the same time period , if you compare the cash values of the two,
you will be amazed by how much Term+Index's cash value will outperform IUL.
Of course, if the policyholder dies during the period, Term+Index's payout... 阅读全帖 |
|
w***n
发帖数: 1519 | 9 Not an expert in bonds, but if I were to invest in bonds, why shouldn't I go
with for example a PIMCO bond fund? Those star managers, more likely than
not, know all the tricks you can think of. Plus, they have enough asset to
diversify and have enough resources to do due diligence and reduce the risks
. A big disadvantage for someone with a small asset size is that, no matter
how many 3% or 5% yield he collects here and there, one default will likely
wipe a big chunk off.
bond |
|
w***n
发帖数: 1519 | 10 This is good suggestion.
Also, I'd suggest for simplicity to just go with "Target 2045" and don't
mess with it, especially if you don't want to learn about investing. A bit
of light on bonds is probably fine IMHO. Bonds are more pricey than equity
at this moment. And they don't give you protection against inflation. Over
the time you just keep saving a chunk of your paycheck and accumulating the
target fund.
Don't watch the fluctuation and don't pretend that you know which direction
the markets ... 阅读全帖 |
|
j****g
发帖数: 597 | 11 finished 5 min ago. writing them down before I forgot.
1. If you have a very large file which consists of many unique tokens(
strings). Easy token is too large to fit in memory. How would you find out
the token that occurs the most in the file.
The hardest question in the interview.
开始我回答用hash function来缩小搜索空间。比如用头两个字符。然后他问what if
the table is too large to fit in memory? 我说那就用swap file, or use OS
memory management related technique, swap in and out the chunk of the hash
table. 后来他忍不住了，问如果给我100台机器 |
|
g*******y
发帖数: 1930 | 12 这个题目挺好的，顶上来大家讨论下。
我觉得用hash来count occurence应该是需要做的吧，有了hash file(may be
distributed across servers)之后就很简单了。
不过对于多servers的大系统如何高效的实现hash？单独的server算自己的hash，再跟
其他server通讯，merge hash file？
有个问题是，因为hash file太大，肯定是要写到disk上的文件，不可能在mem里面装下
，这样就涉及到一个disk write的问题，read时连续地址的item对应在hash表里的地址
都是不连续，而高效的disk write要求一次写入一个large chunk的data，如果你每次
只能disk write一个10Byte的数据，岂不是效率太低了？怎么解决这个问题？ |
|
f*********0
发帖数: 26 | 13 go & relax & eat well
no big chunk of meat, small bite.
just be careful, don't speak when eating ;)
bless~ |
|
n***r
发帖数: 105 | 14 一般来说没有static heap这种说法。不是说data segment吧？
能google出来的就这些了。似乎某些程序会预先跟OS reserve some static memory
heap。
To ensure predictable performance, Flash Lite reserves a fixed-sized chunk
of memory from the operating system (OS) to create what is called a static
memory heap. This memory is reserved when the player starts up and the size
is defined by the device manufacturer.
The OS (based on design by the manufacturer) can optionally provide the
player with additional memory called dynamic memory heap. Flas |
|
r********e
发帖数: 114 | 15 Ali,
I did my homework before interview by studying the job description and my
resume.
And he make it clear that he will give me chunk of time to sell myself.
Therefore, I cover my experience with concise language according to the
requirements of job description.
On the other hand, I'm listening to his reaction. He didn't show any
intention to switch topics or annoying. Therefore, I assume he will like to
understand my expereince better.
So I think it really depends on who is your future manager |
|
b*****n
发帖数: 221 | 16 我的理解是考mapreduce之类的东东.
可以将hash values按range分成1GB的chunks.多次load/save I/O operations. |
|
b********n
发帖数: 609 | 17 第二组也一般，deque明明是queue，什么chunk of memory |
|
s***h
发帖数: 662 | 18
Usually it is, but, not really when your array is 1000 times bigger than
memory. if you use quick select, then you will sometimes need to read the
chunk that you read before, that will cause disk seeking and be expensive.
On the other hand, his heap select only needs to read the whole array
sequentially, this is actually the highest efficiency for disk access.
what do you think? |
|
s******n
发帖数: 3946 | 19 第5题我来写一个吧，每段称为一个chunk
Node* reorder(Node* head, int n) {
Node* newHead = NULL;
Node* chunk_prev=NULL;
Node* chunk_head = head;
while (chunk_head!=NULL) {
Node* saved_chunk_head = chunk_head;
Node* chunk_tail = chunk_head->next;
for (int i=0; i
Node* tmp = chunk_tail->next;
chunk_tail->next=chunk_head;
chunk_head = chunk_tail;
chunk_tail = tmp;
}
if (chunk_prev!=NULL) chunk_prev->next = chunk_head;
else newHead = chunk_head;
chunk_prev = sav... 阅读全帖 |
|
|
l*****a
发帖数: 14598 | 21 if so, how do u push_front and pop_front
Double ended queue
deque (usually pronounced like "deck") is an irregular acronym of double-
ended queue. Double-ended queues are a kind of sequence container. As such,
their elements are ordered following a strict linear sequence.
Deques may be implemented by specific libraries in different ways, but in
all cases they allow for the individual elements to be accessed through
random access iterators, with storage always handled automatically (
expanding an... 阅读全帖 |
|
k***t
发帖数: 276 | 22 自己顶一下，大拿们都不屑于写这个readLine()吗？
这道readLine()的题到底想考什么？
StackOverflow上的一个comment说要考security和memleak。
Implement a function
char* readLine();
which returns single lines from a buffer. To read the buffer, you can makes
use of a function
int read(char* buf, int len)
which fills buf with upto len chars and returns the actual number of chars
filled in. Function readLine can be called as many times as desired. If
there is no valid data or newline terminated string available, it must block
. In order to bloc... 阅读全帖 |
|
d********w
发帖数: 363 | 23 他家就基本上全是老印，但看他们做的还挺high的，blog更新的很频繁
０. hadoop大致问题
partition默认方式是什么，
１. HDFS 讨论，如何设置replica策略，默认是３，放置在local, local rack,
remote,但还是不够efficiently, 我看了mapr他们讲的策略，可以减少replica到１.５
，把数据切成一小块chunk，通过算法可以实现损失任意1/3的数据块，都可以恢复。还
有中想法是讲数据分成hot, cold,对hot数据尽量多写replica，cold就要achieve,定期
做merge
2. High Availability
提到0.23这个版本的改进，变化挺大的
3. next generation mapreduce framework. MR2
大致是把mesos, mapreduce，和其他并行计算的engine，通过配置嵌入到系统中，有
扩展性，方便支持其他的引擎，而不仅仅是以前的mapreduce一种了。
4. Pig组
特别是对join操作的优化，什么skew join, partition join,... 阅读全帖 |
|
d********w
发帖数: 363 | 24 他家就基本上全是老印，但看他们做的还挺high的，blog更新的很频繁
０. hadoop大致问题
partition默认方式是什么，
１. HDFS 讨论，如何设置replica策略，默认是３，放置在local, local rack,
remote,但还是不够efficiently, 我看了mapr他们讲的策略，可以减少replica到１.５
，把数据切成一小块chunk，通过算法可以实现损失任意1/3的数据块，都可以恢复。还
有中想法是讲数据分成hot, cold,对hot数据尽量多写replica，cold就要achieve,定期
做merge
2. High Availability
提到0.23这个版本的改进，变化挺大的
3. next generation mapreduce framework. MR2
大致是把mesos, mapreduce，和其他并行计算的engine，通过配置嵌入到系统中，有
扩展性，方便支持其他的引擎，而不仅仅是以前的mapreduce一种了。
4. Pig组
特别是对join操作的优化，什么skew join, partition join,... 阅读全帖 |
|
z*****n
发帖数: 447 | 25 当时我没有回答出来，interviewer也没告诉具体答案，只说用一种特殊的Hash 函数，
然后用sliding window一个个的滑动计算相邻的chunk的checksum，最后可以比较出来
。结果面挂了。。。：（
我感觉rsync和这个思路很一致，应该就是这样！好像dropbox也是这样做同步的。
时B |
|
q****x
发帖数: 7404 | 26 of course not. r&d 30%~40% on average for it companies.
amzn has retail biz. csr is also a big chunk. so the rate should be lower. |
|
p*******m
发帖数: 47 | 27 我5月初要去Twitter onsite，面的是 Trust & Safety team
2轮电面都比较简单，2个engineers都很nice.
但听说onsite 难度会明显加大, 而且我网上能搜到的它家的题目不多，特别是onsite,
所以心里没底，来版上求经验。
希望有面试经验的同学能和我分享, 如果不方便帖出来，你也可以我站内发信或者
email: d********[email protected]
这是recruiter给我的schedule:
11:30am 去公司签到, 然后面7个人, 其中有个 team Product Manager。每轮45min,
第1轮是 lunch interview,
7轮分别有coding, Ph.D. research presentation, large-scale system design (这
个是open question)
下面是我搜集到的Twitter的题目，有些版上有过了，给后来的同学作参考。
如果你们有新的题, 也可以回在下面.
希望版上它家的信息能多些
++++++++++++++++++++++++++++++++++++... 阅读全帖 |
|
y**********u
发帖数: 6366 | 28 第二题也不难，p2p的东西就太复杂了
简单一点的chunk文件，先传metadata,然后相互穿，类似于broadcast...最后每个机器
自己重组 |
|
v***a
发帖数: 365 | 29 这个不是公网传的，大型server center能用 p2p 么，
一台server一般有好几个张G级网卡连着呢，
核心的机器都是光纤直接连出去的, p2p chunk文件建立协议的的时间都可以传完这个
文件了 |
|
p*g
发帖数: 141 | 30 这一小段程序 还不少bug。。
1. : if (n1==null || n2==null) return null;
2. n1 = n1.next;
n2 = n2.next;
3. 在 p = p.next; 上面一行加了 p.next = new Node(0)
不加的话，到下一个循环p是null
加的话，如果不进入下一个循环， the returned value contains one extra
node at the end.
hehe
Interesting to read such a chunk of code.
n2. |
|
a*****1
发帖数: 314 | 31 why do we use circular link list in place of any balanced binary search tree
in storage allocator? One draw back is that to free() a chunk of memory
allocator has to search the link list and then if found that address then
release , so why not tree to reduce this search and merge?
请高手，给出简单解释。
谢谢 |
|
f*********r
发帖数: 85 | 32 四轮，其中一轮是research:
第一轮：design，设计fb的newsfeed结构，如何储存数据，如何实现给每个用户显示不
同的内容，如何对newsfeed做ranking，如何训练ranking的model，基本上是讨论的形
式，以及讨论各种方式的优缺点。
第二轮：coding，应该都比较typical：（1）给n个2维的点，找出其中离原点最近的k
个。followup：如果n很大，如何做mapreduce；further followup：reduce的时候应该
怎么做最有效？（2）给n个positive int，计算他们两两之间hamming distance的和\
sum_{i
up：数列0到(2^n)-1，计算hamming distance的和（不编程，analytical solution）
；further follow up: 给一棵树，计算每两个节点之间的距离的和，距离定义为path
的长度。
回答：
（1）max heap（最开始说成了min he... 阅读全帖 |
|
c********t
发帖数: 5706 | 33 reduce的时候可以用binary tree的结构reduce，所以如果有m个chunk，就是O(logm)个
reducer。
这里不太懂，为什么不再用一个max heap merge?
写了code，followup解释了用back pointer，然后面试官说不用再写这个了。
准备做一下这个题，follow up用back pointer，是不是还是必须用O(n) space才行？
k
follow
path |
|
x****d
发帖数: 1766 | 34 I did text extraction from someone's PDF transcript file. The original
slides are here.
http://www.slideshare.net/egarbugli/26-time-management-hacks-i-
I find it very good and it is popular. Everybody is sharing with everybody.
So, I did a little work get the text out. You can redo the PPT if you want,
or print it out stick to your screen or your head, haha.
26 Time Management Hacks I Wish I’d Known at 20 by Etienne Garbugli,
Product Design &
Marketing Consultant
1. How I spend my Day How I wish... 阅读全帖 |
|
b****e
发帖数: 365 | 35 转载自Quora:
Recruiting: Have you ever lied on your CV or résumé?
Anonymous
I can attest to Anonymous' answer. One of the best kept secrets in the US
tech industry is the widespread H1b fraud headed by the Indian tech
consulting companies. These companies supply consultants with fake resumes
to all the American companies. These companies displace hundreds or
thousands of American tech workers with inexperienced H1b imports or
unemployed international students. I am an international student who wa... 阅读全帖 |
|
n******e
发帖数: 957 | 36 可不可以像external sort那样，每次shuffle一下每个chunk的头？然后下次shuffle下
个？ |
|
b******7
发帖数: 92 | 37 #define BLOCK_SIZE 1024
#define ALIGN(size,unit) (((size)+(unit-1))/(unit)*(unit))
struct ChunkHead{
ChunkHead * next;
};
#define CHUNK_SIZE ALIGN(sizeof(ChunkHead) + BLOCK_SIZE, sizeof(int))
ChunkHead * firstChunk;
void initChunk(void * addr, size_t size)
{
firstChunk = NULL;
Chunk * pre = NULL;
for(int i = 0; i < size/CHUNK_SIZE; i++)
{
ChunkHead * cur = (ChunkHead *)(addr + i* CHUNK_SIZE);
if( pre != NULL)
... 阅读全帖 |
|
h**o
发帖数: 548 | 38 干嘛不让每个chunk的大小和 1/2 memory一样大 |
|
h**o
发帖数: 548 | 39 干嘛不让每个chunk的大小和 1/2 memory一样大 |
|
g*********e
发帖数: 14401 | 40 不要把整个chunk读进来
读一部分 然后merge 空间用光了就把当前结果写到硬盘上 继续读 merge ... |
|
b*******e
发帖数: 123 | 41 来自主题: JobHunting版 - G新鲜面经 Thanks for sharing..
code for question 1.2
void transform(int A[], int n){
if(n<=1) return;
for(int i = 0 ; i < n-1 ; ++i)
if((i%2=0 && A[i] > A[i+1]) ||(i%2=1 && A[i] < A[i+1]))
std::swap(A[i],A[i+1]);
}
A couple of questions though.
question 2 producer-consumer? producer read files by chunks, push to a queue
, a couple of consumer record statistics and combine result when done?
question 3.1, "random" means 1)any, 2)uniformly distributed, or 3) any
specified index.
Still don't un... 阅读全帖 |
|
I*****a
发帖数: 5425 | 42 http://www.linkedin.com/today/post/article/20131114204433-17970
赚的真多啊
A graduating MBA student in Stanford University’s Graduate School of
Business this year nailed down a private equity job in the northeast with a
total compensation package north of half a million dollars.
This year’s highest paid Stanford MBA, who had a business undergraduate
degree and three to five years of work experience before getting a graduate
degree, reported receiving a guaranteed annual bonus of $337,500–a sum that
d... 阅读全帖 |
|
r****m
发帖数: 70 | 43 9月份的面试，连续四天面了LGTF，准备面试的半年多时间来从本版受益匪浅，现在把
面经写出来回馈本版，希望大家把好的传统延续下去。
L偏重设计，也可能与面的组是platform有关，6个面试有三个是设计，而且涉及很多细
节，比如index，distribute hash, circule counting. 有一面是manager问项目，个
人觉得选一个自己从头到尾做过的项目，然后按我下面的6点进行准备，基本就够了。
L是有题库的，建议多刷版面和glassdoor。
G偏重coding，每一面都是coding开始，而且占很大比例，如果时间多的话可能有两个
coding，也有可能接一个design问题。
T的面试最没规律，感觉基本是面试官自己决定问什么，所以这里不怎么好做总结。
F的面试是最标准化的，两个半coding + 一个design + 半个项目介绍 (项目介绍同上
面L的), F的题目重现率比较高，看版上的题目就差不多了，design问题基本在之前版
上归纳的几个类别： 设计feed，message, search，存储，都和大数据沾边。
LFT面试官大部分是同胞，大部分同胞是... 阅读全帖 |
|
a********9
发帖数: 129 | 44 楼主能讲讲L的另外两道design题么， 感谢！
3. Design a Message store system (in-memory storage) [seq_id, len, data]
chunk
看起来像memcached?
4. Design monitoring system, circular array, storage, aggregation
是指这个么http://en.wikipedia.org/wiki/System_monitoring？怎么用circular array? |
|
r****m
发帖数: 70 | 45 9月份的面试，连续四天面了LGTF，准备面试的半年多时间来从本版受益匪浅，现在把
面经写出来回馈本版，希望大家把好的传统延续下去。
L偏重设计，也可能与面的组是platform有关，6个面试有三个是设计，而且涉及很多细
节，比如index，distribute hash, circule counting. 有一面是manager问项目，个
人觉得选一个自己从头到尾做过的项目，然后按我下面的6点进行准备，基本就够了。
L是有题库的，建议多刷版面和glassdoor。
G偏重coding，每一面都是coding开始，而且占很大比例，如果时间多的话可能有两个
coding，也有可能接一个design问题。
T的面试最没规律，感觉基本是面试官自己决定问什么，所以这里不怎么好做总结。
F的面试是最标准化的，两个半coding + 一个design + 半个项目介绍 (项目介绍同上
面L的), F的题目重现率比较高，看版上的题目就差不多了，design问题基本在之前版
上归纳的几个类别： 设计feed，message, search，存储，都和大数据沾边。
LFT面试官大部分是同胞，大部分同胞是... 阅读全帖 |
|
a********9
发帖数: 129 | 46 楼主能讲讲L的另外两道design题么， 感谢！
3. Design a Message store system (in-memory storage) [seq_id, len, data]
chunk
看起来像memcached?
4. Design monitoring system, circular array, storage, aggregation
是指这个么http://en.wikipedia.org/wiki/System_monitoring？怎么用circular array? |
|
n********e
发帖数: 41 | 47 楼主 看到 你 isDiffByOne 函数就知道你必然要超时
Leetcode免费给你用 就别挑剔了。凡事要先从自身找问题
你那个测试不是大数据
真正的大数据在这里：
start = "nanny";
end = "aloud";
String[] d = {"ricky","grind","cubic","panic","lover","farce","gofer
","sales","flint","omens","lipid","briny","cloth","anted","slime","oaten","
harsh","touts","stoop","cabal","lazed","elton","skunk","nicer","pesky","
kusch","bused","kinda","tunis","enjoy","aches","prowl","babar","rooms","
burst","slush","pines","urine","pinky","bayed","mania","light","flare","
wares","wom... 阅读全帖 |
|
o**********e
发帖数: 18403 | 48 Bangalore: Every day the Indian IT industry wakes up with new challenges
which could be from hiring skilled professionals to cost cutting during
recession. Fake resumes are one of them. Since this is the first step in
hiring, it makes it all the more difficult for companies to pick the right
candidate. Despite having large HR departments, many IT companies have
failed in identifying fake resumes. According to Employee verifications
firms, recruitment companies and HR heads in the IT sector feel ... 阅读全帖 |
|
o**********e
发帖数: 18403 | 49 Google like other big tech companies has huge R&D centers in India,
including a big chunk of their Enterprise team (not sure why, but because
they want to be far away from their customers and meet their customers
midnight?). These Indian sites are famous for stealing credits and projects
from hard working teams all over the rest of the world. Google is now
slapped with extortion racket by the Indian government. Maybe Google forgot
to grease some bureacrat's palms?
http://www.eweek.com/enterprise... 阅读全帖 |
|
z*********5
发帖数: 3 | 50 最近在准备面试刷版上面经，下面几道题想了好久不知道应该如何考虑，希望版上大牛
们可以不吝赐教。。。。
1.Design a Message store system (in-memory storage) [seq_id, len, data]
chunk， 原帖地址：http://www.mitbbs.com/article_t/JobHunting/32578885.html
这题不知道应该注意的是什么？开一个循环数组+hashmap的话可以按seq_id 得到data,
maintain数组头和尾，内存满的时候把最早的覆盖。 如果还要求得到seq_id是某个区
间内的data,是不是可以考虑用类似于B Tree建index的方式来按顺序遍历各个seq_id?
2.Multiple threads can publish and receive each other's message: whenever a
thread publishes a message, all the other threads can receive and print out
that mess... 阅读全帖 |
|
