In SGD fitting, does maxiter=100 mean exactly one hundred single-row gradient updates?
Classic MLLevel: FoundationEpoch versus update budget
In SGD fitting, does max_iter=100 mean exactly one hundred single-row gradient updates?
Hints
- Distinguish an epoch from a sample update.
- A cap is not a promise to use the full budget.
Answer
No. max_iter caps passes over the training data, and each epoch contains many sample updates. Early convergence can stop before the cap.
Source and licence: scikit-learn 1.7.2 — sgd · BSD-3-Clause · answer: Quantoros editorial