Deep Learning (DL) · Intermediate level

Intermediate level · Module 4: GPUs and Efficient Training · Lesson 3 of 3

Planning Experiments within a Budget

Plan and track experiments so limited compute time produces the most learning.

In this lesson you will

  • Prioritise experiments by expected value
  • Track runs and stop unpromising ones early

Key concepts

experiment plancompute budgetearly stopping of runsexperiment tracking

The lesson

Compute is limited and may cost money. List candidate experiments, estimate their cost and expected benefit, and run the most informative ones first.

Start with short runs on subsets to compare options, then train the best ones fully. Stop runs that are clearly worse early.

Track every run's settings and results; without records, expensive runs are repeated by mistake.

Step by step

  1. List candidate experiments with estimated GPU time.
  2. Prioritise them within a fixed budget.
  3. Run short comparison runs.
  4. Record all runs in a tracking table or tool.

Practice activity

Write and follow an experiment plan for the project within the given GPU budget.

Expected result: A plan, a run log and a short reflection on what was learned per GPU hour.

Check your understanding

  1. Why run short experiments on subsets first?

    Show answer

    To compare options cheaply before spending time on full runs.

  2. Why track every run?

    Show answer

    To compare results and avoid repeating expensive runs.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.