Stochastic gradient descent on the How to Train Models track. SGD estimates the gradient on a batch and steps that way. Noise can help escape sharp minima. The learning rate still dominates.
This lesson assumes you already worked through What an optimizer does.
The idea in practice
Use momentum if the loss zigzags. Plot the loss every N steps, not only every epoch.
A concrete check
goal = {
'track': 'How to Train Models',
'lesson': 'Stochastic gradient descent',
}
checks = [
'input available at decision time',
'score matches the real decision',
'failure case written down',
]
print(goal['lesson'])
for item in checks:
print('-', item)
Run the sketch locally if you have Python. The printout is a reminder of the checks, not a trained model. Replace the strings with the real inputs from your own example before you treat it as a design.
What usually goes wrong
A learning rate so high the loss becomes NaN, then switching optimizers instead of shrinking the rate. When this happens, stop adding parameters or tools. Fix the check, the data, or the permission, then run the same example again.
What to write down
- The input you are allowed to use at decision time.
- The output and the score or pass rule.
- One failure you will test on purpose.
- What you will not claim the system can do.
Practice
Describe a healthy SGD loss curve versus a diverging one.
Self-check
- Say Stochastic gradient descent in one sentence that mentions an input and an output.
- Name the failure mode in this lesson and the check that would catch it.
Done when: you can explain this lesson without the page open, and you have a written failure case.