Track
train-models
How to Train Models
36 lessons: objectives, splits, optimizers, curves, checkpoints, and when to stop—plus 108 MCQs.
- Mode
- none
- Practice
- Read / quiz
- Lessons
- 36 units
Before you start
The training loop in practice: objective, frozen split, loss, optimizer, curves, checkpoints, and the rule for stopping.
Most failed trainings are split bugs, wrong losses, or runs nobody can reload—not a shortage of layers.
Notebooks, training jobs, and model reviews where someone must defend a checkpoint.
Follow the checks in each lesson on a small dataset. Save artifacts. Score validation before you look at a final test.
After /ai-learning/intro and a working Python install—when you will actually fit a model.
Lesson order
Sequential — follow top to bottom36 lessons are live in this track. Start from step 01 for the smoothest path.
-
01 intro What training actually changes
beginner
Open → -
02 objective Choose the objective first
beginner
Open → -
03 dataset-split Freeze the split
beginner
Open → -
04 preprocessing Preprocessing that fits on train only
beginner
Open → -
05 baseline-model Train a baseline before a fancy model
beginner
Open → -
06 loss-functions Pick a loss you can explain
beginner
Open → -
07 optimizers What an optimizer does
beginner
Open → -
08 sgd Stochastic gradient descent
intermediate
Open → -
09 adam Adam and when it hides problems
intermediate
Open → -
10 batch-size Batch size and memory
intermediate
Open → -
11 schedules Learning-rate schedules
intermediate
Open → -
12 initialization Initialization
intermediate
Open → -
13 normalization Normalization layers
intermediate
Open → -
14 regularization-train Regularization during training
intermediate
Open → -
15 checkpoints Checkpoints
intermediate
Open → -
16 early-stop-train Stopping on validation
intermediate
Open → -
17 class-imbalance Imbalanced classes
intermediate
Open → -
18 augmentation Augmentation
intermediate
Open → -
19 validation-during-training Validation during the run
intermediate
Open → -
20 overfitting-signals Reading the curves
intermediate
Open → -
21 gradient-problems Vanishing and exploding gradients
advanced
Open → -
22 mixed-precision Mixed precision
advanced
Open → -
23 gpus Using a GPU without fooling yourself
advanced
Open → -
24 small-data Training when data is small
advanced
Open → -
25 hyperparameter-search Searching hyperparameters
advanced
Open → -
26 experiment-tracking Tracking runs
advanced
Open → -
27 seeds Seeds and repeated runs
advanced
Open → -
28 debugging-training Debugging a run that will not learn
advanced
Open → -
29 evaluation-after-train Evaluate after you stop
advanced
Open → -
30 calibration Calibration
advanced
Open → -
31 model-selection Selecting the model to keep
advanced
Open → -
32 saving-artifacts Artifacts you must save
advanced
Open → -
33 retraining Retraining later
advanced
Open → -
34 cost-of-training The cost of a run
advanced
Open → -
35 interview Interview review
advanced
Open → -
36 capstone Train and document one model
advanced
Open →