MobyDicks
A working notebook of ML ideas, small experiments, and questions worth chasing.
Ongoing ideas and thought experiments · Python / Jupyter
MobyDicks is where I explore machine-learning ideas before they become a paper or a standalone project. Most entries are small notebooks: a question, a quick experiment, and sometimes a useful finding or an unfinished direction. The name follows Ahab’s fixation in Moby Dick—ideas that keep asking to be investigated.
What’s inside
- Graphs: representing tabular data as graphs, multi-hop node classification, test-time training, and transfer learning with node and relation representations.
- Features and geometry: PCA-based feature importance, random projections, signed distances, and geometric checks.
- Learning: anomaly detection, positive-unlabeled learning, tree embeddings, and alternative boosting ideas.
These are exploratory notebooks with varying levels of completeness. Their value is in the questions and experiments; they are not a collection of validated research results or a maintained software package.
Explore or collaborate
Browse the notebooks on GitHub, starting with the Graph, Features, or Learning folders.
If an idea connects with your work, get in touch. I am happy to discuss possible extensions or collaborations.