Exercise for lecture 2 - Data augmentation

Learning Objectives

• Explore data
• Augment data
• Train a simple model and evaluate the effect of using augmented data

Preparation

• Accept assignment: https://classroom.github.com/a/Zz9PEnY2
• Clone your student repository (git clone)
• Run uv sync and check everything is correct with uv run hello.py
• cd exercise
• Unzip 02_Files.zip
• Start Jupyter

Exercise

1. Evaluate the nearest neighbour baseline "properly" in this notebook. Complete the functions in tasks.py and pass the tests. You will have to:
2. Create a subset of the original datasets with 500 images. Create an augmented data set of 2500 images from the selected subset. Pay attention to obtaining a representative balance between healthy and malign samples. The augmented images should be of size 64,64.
3. Fit a KNNs classifier using
   1. the original subset of 500 data samples. You should achieve >60% accuracy.
   2. the augmented dataset. You should achieve >70% accuracy.

   Note: You will have to play with the number, type and hyperparameters of the augmentations and the kNN classifier. Note: You will have to implement the functions to train, predict and evaluate the kNN model.

4. Compare the performance using the confusion matrix. Plot it on this notebook.

Tip: Solve each of the tasks first on the notebook, so it is easier to see the input and output of the functions. Check the file test_exercise2.py and the docstrings of each function to get more information on how to implement them.