CHAPTER 19 / 40
Numerical Methods for ML
3 lessons · Explore, experiment and understand.
DOJO PETYour little coding companion
DOJO / YOUR GUIDE
Let’s figure it out.
Ask me about this course or where to start.
Chapter lessons
3 lessons
Why this matters in ML
Floating-point reality, condition numbers and the log-sum-exp trick are what stand between your theory and a NaN — every production ML pipeline relies on them.
- stable softmax
- mixed precision
- well-conditioned linear systems
- Step 86~10m12
Floating-Point Reality
Computers can't represent most real numbers exactly. Knowing the shape of float32 — and what breaks when you ignore it — is the difference b…
Open lesson - Step 87~10m3D · 112★Most cinematic
Condition Numbers and Numerical Stability
Some linear systems amplify small input errors into huge output errors. The condition number quantifies this amplification — and explains wh…
Open lesson - Step 88~10m12
The Log-Sum-Exp Trick
One of the most used tricks in ML: rewrite log(Σ exp(x_i)) so it never overflows. It's the engine behind every stable softmax, log-likelihoo…
Open lesson