Objective: turn a symptom into testable hypotheses.
Plot train and validation loss per epoch, then the business metric. Do not conclude on a single epoch: look at trend, variance and time of the best checkpoint.
| Observation | Hypotheses | Targeted actions |
|---|---|---|
| train and val both poor | underfitting, bug, weak signal | baseline, capacity, data |
| train good, val bad | overfitting or shift | regularize, audit the split |
NaN losses | instability, invalid data | lower lr, inspect values |
| val better than train | dropout/augmentation, possible leak | compare modes and pipeline |
Before a long training run, ask the model to memorize 16 to 64 examples. If it cannot, suspect an error in shapes, labels, loss, gradients or capacity. Passing this test does not prove generalization; it only proves the pipeline can learn.
Preview — the rest of the lesson is for enrolled readers.
Your access is tied to your account, not to this link. Sign in with the same email you used in class: your course is waiting, no need to enter the code again.
Sign inNo account yet? Create oneThe first modules of the course are open to everyone. For the rest you have three options: buy this course once and for all, subscribe, or enter the code handed out in class.