M1 Artificial Intelligence · semester 7 · Machine learning
Machine learning
Seven labs in scikit-learn and PyTorch, from a first confusion matrix to a fine-tuned ResNet and a recurrent network reading a building's sensors.
The labs
Lab1
Performance evaluation
Six predictors on breast-cancer data, compared by accuracy, precision, recall, F1, ROC and AUC.
Lab2
Supervised learning and SVMs
A linear SVM on synthetic blobs, then avalanche prediction from weather and snowpack data, tuned through the soft-margin C.
Lab3
Unsupervised learning
k-means on digit images, then PCA and t-SNE, read against clustering agreement and silhouette scores.
Lab4
Regularisation
Predicting IBD status for 396 patients from 1,939 gut-microbe abundances — far more features than patients, so regularisation carries the model.
Lab5
Deep learning
Hitters' salaries by linear regression and Lasso, then a two-layer network in PyTorch with a hand-written training loop.
Lab6
Fine-tuning
A ResNet18 pretrained on ImageNet, fine-tuned for object detection on Pascal VOC images.
Lab7
Recurrent networks
Forecasting humidity from the CUBEMS building data with an RNN and an LSTM, windows of past steps predicting the steps ahead.
Reading it
Source
The folder on GitHub
Seven lab folders, each with its completed notebook once the work is done.
README
The course README
The full table of labs across both halves of Introduction to AI, what each is solved with, and what is provided going in.
The lab subjects, the official corrections, the helper modules and the datasets are not redistributed here. Each lab keeps them on disk, out of the repository; what is committed is the work written against them.