Part 2 · Roadmap: Who to Learn First, How Long to Invest, How to Verify, and the Degree at the Finish Line

I. You Have the Map and the Ammunition — Now You Need the Route

The first two posts laid out the coordinate system (what to learn and why) and the courseware map (where to find resources for each course). This is the final piece of the puzzle: how to proceed — who comes first, who comes second, how long to invest per course, what counts as proof you’ve learned it, where people most often get stuck, and finally, how the full “self-taught → degree” pipeline lands.

II. The Main Track: 18–24 Months, Six Stages

Designed for 15–20 hours per week (part-time self-study). One course at a time — don’t bite off more than you can chew:

1
2
3
4
5
6
7

Months 1–2      CS106A → CS106B (Intro to Programming)
Months 3–4      Linear Algebra (18.06) ∥ CS103 (Discrete Math)
Months 5–6      CS107 (Computer Systems) + CS109 (Probability)
Months 7–8      CS161 (Algorithms)
Months 9–12     CS229 (Machine Learning) → First Project
Months 13–16    CS224N or CS231N → Second Project
After that      Electives + Continuous Projects

Why this order: each stage builds directly on the one before it. Without programming intuition, even rock-solid math won’t translate into anything you can actually use. Without math, algorithms and machine learning are castles in the air. Without algorithms, AI courses are all “vibes” and zero derivation. Two notes: first, CS106A now teaches Python (switched from Java in 2023); second, CS103 Discrete Math has no prerequisites, so you don’t need to put it before programming — pairing it with CS106B or dropping it after CS107 both work. The real entry point is always: write code first.

III. How Long to Invest Per Course (Magnitude)

These are part-time figures at 15–20 hours per week. Don’t compare your pace to full-time graduates; what matters is consistency.

IV. How to Verify Each Course (What You Can Actually Build)

Stop yourself from thinking you’ve learned something when you haven’t — here’s an objective benchmark per course:

• After CS106B: independently build a 500+ line mini-project (e.g., a CLI tool or a small game)
• After CS107: read and understand the labs in CSAPP, explain the full journey “from source code to running program”
• After CS161: solve LeetCode Medium problems independently, explain when to use dynamic programming vs. graph algorithms
• After CS229: implement a classic model from scratch (linear/logistic regression, neural network) and run it on a real dataset
• After CS224N/CS231N: build your own demo (text classifier / image recognition app)

If you can’t meet the bar, don’t rush into the next course. Verification is the only defense against fooling yourself.

V. Five Pits Where People Get Stuck

1. Overloading — taking three courses at once — and ending up finishing zero. One at a time. Finish a course before starting the next.

2. Skipping math — halfway through CS229 you realize linear algebra and probability are blocking you, so you backtrack. Double the wasted time. Math is the foundation; you can’t skip it.

3. Only watching lectures, never coding — CS is a craft. Ten viewings of a video are worth less than one run of you typing the code yourself. Do every assignment on your own, even if it’s slow.

4. Quitting when videos are inaccessible — Stanford hosts some course videos on campus servers that aren’t publicly reachable. Don’t get stuck: switch to the corresponding MIT course videos flagged in the courseware map. Same material, different source.

5. Not keeping records — open a dedicated note repo for every course. What you leave behind is your first-hand portfolio for future interviews, freelance work, and monetization.

VI. Your Acceleration Points (R Background)

Let’s be honest: your background comes with built-in advantages. You know R and you have statistical thinking. That means:

• Probability (CS109) can be fast-tracked — you already have the stats intuition; focus on the key parts
• The math derivations in machine learning — the linear algebra + probability sections will save you 1–2 months compared to a complete beginner
• The real hard啃 pieces are threefold: programming intuition (CS106A/B), systems (CS107), and algorithms (CS161). These have no shortcuts — everyone hits the wall here, and you should expect the same

VII. The Finish Line: How the Degree Actually Lands

Finishing the courses isn’t the end — the degree is what makes all this capability “real.” The full pipeline:

1
2
3
4

Self-study 18–24 months (this series)
  → Earn GT edX Verified Certificates (Python / Java OOP / Data Structures & Algorithms, with exams)
  → Apply to OMSCS (Georgia Tech Online MS in Computer Science, two intakes per year: Fall / Spring)
  → Work while studying, 1–2 courses per semester, graduate in 3–6 years

Key numbers (verified as of 2026):

• Total OMSCS tuition is roughly $7,000–$9,500 USD (30 credits × ~$227/credit + per-semester fees; total varies with your pacing; summer sessions offer price breaks). This is the highest-value top-tier CS master’s anywhere in the world.
• No GRE required. Non-CS majors follow the official prerequisite path (evaluated case by case — don’t expect leniency).
• GT edX certificate budget: ~4,000–5,000 RMB (three courses, including exams)
• Total budget: ~80,000–100,000 RMB, paid year by year — not a lump sum
• International student hard threshold: TOEFL ≥ 100 or IELTS ≥ 7.0; complete 2 foundation courses within 12 months of enrollment with grades of B or higher
• CSCSE certification note: Fully online degrees cannot be authenticated by China’s Ministry of Education service center. Not recognized for civil service exams, state-owned enterprise hiring, or university tenure tracks. Recognized by private companies, multinationals, startups, and overseas employers.

The single most important reminder (said again): free MOOCs alone won’t cut it for your application. OMSCS explicitly states “MOOCs alone are rarely sufficient unless you hold a STEM undergraduate degree.” So the certificate courses are an indispensable link in this degree pipeline — free courses build knowledge, certificates prove it, and the degree makes it stick. All three are required, in the order laid out above. Also note academic integrity: don’t use paid answer sites (Chegg, My.Answers, and the like). OMSCS uses MOSS for plagiarism detection — a violation gets you dismissed instantly, and there are also copyright risks back in China.

VIII. Closing: A Complete Self-Study System

By now, these three posts together form a complete system:

• Coordinate System (Part 1): know what to learn and why
• Courseware Map (Part 2): know where the resources for every course live
• Roadmap (this post): know how to proceed, how long it takes, how you’ll know you’re done, and what the finish line looks like

The map, the ammunition, the route, the destination — all four are in your hands. The only thing left is: start walking.

Start your first course today, and 16 months from now you’ll be standing somewhere entirely different. I chose to lay this map out because it deserves to be taken seriously. The rest is up to you to walk it yourself.