Module 01 Activity

Scenario

Before the first model, establish what your machine can do and what you will record about every run.

Task

1. Run the setup check and note your torch version and available device.
2. Reproduce the XOR comparison. Record the accuracy of the linear model and of the one-hidden-layer model.
3. Re-run the capacity demonstration at widths 2, 64 and 512 and record training loss for each.
4. Run the same script twice without a seed, then twice with one. Record whether the numbers match.
5. Write a five-line evidence note for one run.

Deliverable

An environment note: torch version, device, and the five figures from steps 2-4, plus your evidence note.

Check your work

• The linear model on XOR should sit at 0.5 — chance. One hidden layer reaches 1.0.
• Training loss on random targets falls towards zero as width grows. The 512-unit model has more

parameters than training rows, so it can memorise them.

• Unseeded runs differ; seeded runs are identical to the last decimal.

What good looks like

Your note lets someone else reproduce a number. If it omits the seed, it does not.

Mark complete