A short homework — get the environment ready and decide what your network will recognize.

1. Run python check_setup.py. Make sure you see the PyTorch version and a line "Using CPU" or "Using CUDA". If you got ModuleNotFoundError, install PyTorch (pip install torch) and run it again.
2. Pick your categories. Decide which classes your classifier will recognize — whatever you like, from your own domain. For a start, 3–5 clearly different categories work well.

Good examples:

• Pets — cat, dog, rabbit
• Fruit — apple, banana, orange, grape
• Transport — car, bicycle, bus

Avoid at the start: very similar classes (husky vs malamute) or too many categories (20+) — it gets harder and slower. You can push further later.

To submit: the check_setup.py output (version + CPU/GPU) and your list of 3–5 categories in one sentence.

Tip. The environment works and you have your categories — the foundation is laid, and the accuracy narrative begins. Next lesson we take on tensors: how an image becomes a grid of numbers PyTorch understands. That is where the real code starts — but one step at a time, as always.