GT

交了paper之后好像工作效率一直很低下，最近两周貌似也没干什么。打算在一篇paper上打打酱油，然后要思考下一步要做的事情。

最近看了很多paper，产生了一些稀奇古怪的想法，不知道这些想法靠不靠谱。感觉思考的还不够深入，可能我应该像写introduction一样把

一件很让人郁闷的事情是，自己想的东西发现已经有人做过了.. 不过这件事大概太常见了吧。从好的角度来说，这至少说明我的这些idea是靠谱的（因为别人做出来了并且发了paper）？

现在的目标是在学期结束前做一个新的project并且投到某个地方（也许是CIKM）。另外希望under submission的两篇文章不要被批的太惨...

Weisfeiler-Lehman Graph Kernels: http://www.jmlr.org/papers/volume12/shervashidze11a/shervashidze11a.pdf

Graph Kernels:
http://www.jmlr.org/papers/volume11/vishwanathan10a/vishwanathan10a.pdf

**Deep Graph Kernels:
https://users.soe.ucsc.edu/~vishy/pubs/YanVis15.pdf

subgraph2vec:
https://arxiv.org/pdf/1606.08928.pdf

http://media.cs.tsinghua.edu.cn/~multimedia/cuipeng/papers/Network%20Representation-Tutorial.pdf

I am back!

Link to the contest: http://codeforces.com/contest/732

Problem C:
Brute-force is enough for this one. Enumerate the 4 configurations for the first day, and the 4 four configurations for the last day. For example, the 4 configurations for the first day: arrive before breakfast, before dinner, before supper or after supper.

After that, assume Vasiliy have $a$ breakfasts, $b$ dinners and $c$ suppers on the other days -- then he needs to stay for $max(a, b, c)$ additional days to finish that many meals. Choose the best configuration from the $4 * 4$ different configurations.

Problem D:
Binary search + Greedy. One observation: if we could pass the exam for course $i$ on $k$ different dates, choose the last date is always the best, so that we could prepare on some other course earlier.

Thus, we could do binary search on the answer, and judge if it is possible to pass all exams.

Problem E:
The most important thing: each socket can only connect to at most ONE computer.

Sort the power of sockets in ascending order. Then for each socket $i$ with power $s_i$, check if there is a computer with power $s_i, s_i / 2, s_i / 2 / 2, \cdots, 1$. You must check in this order to ensure minimum number of adapters are used.

Get TLE using Python, faint. I will try C++ later... -- Got accepted with C++.