usaco-guide
usaco-guide copied to clipboard
cpinitiative
Mixed Problem Sets
Will mixed problem sets be helpful / is it within the scope of the USACO Guide?
Motivation
Currently, the USACO Guide has a reasonable number of problems per topic, but no mixed problem sets. Therefore, people might say "I went through the USACO Guide, did all the problems, and still can't promote." In this situation, will providing very large mixed problem sets for practice help?
The goal would be to make the USACO Guide an actual one-stop-shop. People go through the USACO Guide to learn all the topics, and then they do practice problems until they promote. Currently I think the Guide does a good job of the former, but after you've gone through the Guide, if you're still unable to promote you basically have to figure out how to get problems yourself (especially once you've exhausted problems from the Guide, which honestly aren't that many?).
Getting Mixed Problemsets
Basically for every problem we'll need (not all are necessary):
- Basic problem info (name, URL, etc)
- Division + difficulty relative to division
- Tags
- Solutions
I don't particularly know how to get this information, especially for a lot of problems.
We could either crowdsource from users or find problems ourselves (idk how long this would take).
Really old USACO problems might be a good source? We have the problem statements, test data, and solutions, and we can relatively easily create an online judge to make all those problems submittable. I don't know how good the quality is though.
Using Mixed Problemsets
Some random ideas:
- After every few modules, list a bunch of problems from the preceding modules
- At the end of every division ("additional practice"), put a bunch of problems (we already have some, but like, a lot more)
- On the problems page, make a better "practice problemset generator" (filter by topic, difficulty, remove problems you've already solved, etc)
- Make a "mock practice contest generator"
i can probably make a CLI for that integrating it into the website? yeah that's beyond me all for the idea though
I think crowdsourcing is best, w/ maybe some rating system on problem quality. If difficulty doesn't matter, we can also just scrape them from somewhere
Really old USACO problems might be a good source
maybe training pages as well?
Regarding the comment above, difficulty rating is helpful, but I think a quality rating is not. Doing the problems in increasing order of difficulty makes sense, but doing the problems in increasing or decreasing order of quality doesn't. If we think a problem is of questionable quality then we shouldn't include it. Also, problem quality is more subjective than difficulty.
As I mentioned to Nathan previously I don't see how any problemset we create would be more effective than doing problems from an OJ that supports sorting by solve count or difficulty (e.g., https://csacademy.com/contest/archive/).
Some recent related feedback:
I think a bigger list of problems for each section would go a long way. This year, after finishing almost all of the guide for bronze and doing around double that amount in other bronze problems as well as codeforces, I still was unable to promote due to an extremely hard problem 3. I think in general, lots of contests trend towards 2 easy problems and 1 hard one. I think while usacoguide is great for easy problems with clear algorithms, we need more for the harder ones on the more ad hoc/greedy algorithms.
Given the influx of more observation type problems, maybe a dedicated section in the silver division where problems are pulled in from codeforces. Also, giving similar CF extra problems to recent silver problems that use similar intuition.
I want a larger section of problems that haven't been categorized
My opinion hasn't changed; if Codeforces problems are helpful for USACO Bronze or Silver then just go do them, what do you need the USACO Guide for? I'm not aware of codeforces problems that are more similar to current USACO problems than past USACO problems.
people might be too lazy to go & find problems themselves on cf, the problemset is just one giant hodgepodge of problems, all the way from bronze to plat you can filter by tag/rating, but knowing the tags of a problem can go a long way in helping you solve it
Just filtering by rating should suffice? Assuming you want uncategorized problems.
i'm not sure, but curating a list of problems might encourage people to do them more than just "look at cf lmao" again, i'm not agreeing with them, just trying to think of their possible reasons