How I Prepared for ICPC Pakistan Nationals

In November 2024, our team from NUST placed 3rd at ICPC Pakistan Nationals. Here's an honest breakdown of what the preparation looked like.

Team Structure

We were a team of three: myself (primary on graphs + number theory), a teammate strong in geometry and DP, and a third who handled implementation-heavy problems under time pressure. Knowing your role matters as much as raw problem-solving ability.

The Problem Volume Question

A common piece of advice is to grind 500+ problems before ICPC. I think this is wrong. What matters is the distribution of problems, not the count.

The topics that showed up most reliably in Pakistan regionals:

Graphs: Shortest paths, SCC, articulation points, min-cut
DP: Interval DP, bitmask DP, digit DP
Data structures: Segment trees with lazy propagation, BIT, sparse table
Number theory: Sieve variants, modular arithmetic, CRT
Strings: KMP, Z-algorithm (suffix arrays rarely appear)

Geometry is a trap for Pakistan regionals — usually one problem, often skippable.

Practice Structure

Months 1–3: Fundamentals

I used Codeforces to target specific topics. For each topic, I'd:

Solve 5–8 problems at difficulty 1700–2000
Upsolve anything I couldn't solve in 45 minutes
Write a template for the algorithm

The template discipline is underrated. If your segment tree template has a bug, you want to find it in practice, not in contest.

Months 4–5: Virtual Contests

We started running 3-hour virtual contests on Codeforces together, 3× per week. The coordination practice is crucial — who types, who reads problem statements first, how you communicate.

A rule we adopted: the person not coding reads the next problem. You should always be ahead of your current implementation.

Month 6: Past ICPC Problems

ICPC problem sets are public. We solved the last 5 years of Pakistan divisional problems. Recurring themes showed up: flow problems (almost always one), and a constructive that looks hard but isn't.

Contest Day

The first 15 minutes are critical. We split all problems by person, skim for easy ones, and have the two easiest coded within 20 minutes. This builds momentum and balloon count.

We ended up solving 6/11 problems. The 7th — a DP on trees — we had a correct approach but ran out of time getting the implementation clean.

What I'd Do Differently

Invest more time in interactive problems. They appeared in more recent ICPC sets and we had almost no practice with them.

Also: practice typing faster. Sounds trivial; isn't.