OMSCS CS7646 (Machine Learning for Trading) Review and Tips
Eugene Yan
A grad student just wrapped up Georgia Tech's OMSCS Machine Learning for Trading course and posted a detailed review. It's a rare inside look at how ML gets taught for real-world stock trading, not just theory.
Based on reporting by Eugene Yan — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Eugene Yan, an OMSCS student with a background in data science across HR, e-commerce, and healthcare, spent his spring 2019 term grinding through CS7646, Georgia Tech's Machine Learning for Trading course, and came out the other side with a genuinely useful breakdown for anyone considering it.
The course leans heavily on eight coding assignments, roughly one every two weeks, ranging from analyzing the Martingale betting strategy to building decision tree learners, Q-learning robots, and eventually a full trading strategy learner. Some projects took a couple hours; others, especially the ones asking students to frame market data as a supervised learning problem, ate up 10 to 20 hours including report writing. Two exams, 30 multiple-choice questions each in 35 minutes, rounded things out. Yan reports the OMSCentral crowd rates the class 4.3 out of 5 with a difficulty of 2.5 out of 5, and pegs weekly workload around 10 to 11 hours, light enough to pair with a second course.
What stands out in his account isn't the grading logistics, though he does praise the fast, largely automated turnaround, it's the framing problem at the heart of applying ML to markets. Should the model predict tomorrow's price, a regression task, or just whether to buy or sell, a classification task? That question, Yan notes, rarely comes up in standard machine learning courses but sits at the center of this one. He picked up practical exposure to financial instruments like options, technical indicators such as Bollinger Bands and MACD, and reinforcement learning applied directly to trading decisions.
The course also underwent a leadership change worth mentioning: Professor David Joyner took over after JPMorgan hired away original instructor Tucker Balch, a detail Yan treats as proof the material has real industry pull. Still, after watching his own trading algorithms flounder on out-of-sample data, Yan isn't rushing to bet real money on anything he built. He's walking away instead with sharper instincts for sequential modeling that he plans to apply back at his healthcare job, and a renewed interest in managing his own investment portfolio using the fundamentals the course reinforced.
My take — AI-written commentary, not fact-checked reporting
I like seeing course reviews like this treated as real content, because the gap between 'we taught an ML model to trade stocks' and 'this model actually works out-of-sample' is exactly the gap the entire AI industry conveniently glosses over in its marketing. Yan's honesty about not trusting his own algorithm with real money is the most useful data point in the whole piece. More builders should say that part out loud.
Read more about this at: Eugene Yan