Read it, then run it

This page is the read-only chapter intro. To actually execute the cells (load the dataset, train a model, render GradCAM heatmaps on your own images), open the Colab notebook in a free T4 runtime. The curriculum is designed to be executed, not read.

Chapter 0: Environment Setup

Learning objectives: - Install and verify all required Python packages - Confirm GPU access on Google Colab - Set random seeds for reproducible results

Clinical context: Before running any experiment in the lab, you check your equipment is calibrated and working. This chapter is the computational equivalent - making sure our tools are ready and our results will be reproducible.

Why reproducibility matters

In clinical AI research, reproducibility is not optional - regulatory bodies such as the FDA require that your results can be independently verified. If another team runs your code and gets different numbers, your findings cannot be trusted for clinical decision-making.

We ensure reproducibility by: 1. Pinning package versions - so the same code runs identically months later 2. Setting random seeds - so random operations (data shuffling, weight initialization) produce the same sequence every time 3. Using deterministic GPU operations - at a small cost to speed, this removes GPU-level randomness

Think of it like a clinical trial: every step of the protocol is documented so that the study can be replicated at another institution. The same principle applies to computational experiments.