This codebase was created as part of a replication study concerning Conditional Neural Processes. It contains implementations of Latent and Conditional Neural Processes, Attentive Neural Processes, and Convolutional Conditional Neural Processes. The work was conducted as part of the MLMI MPhil at the University of Cambridge.
We give a brief overview of the contents of the different repository folders:
- data/ contains classes to generate the 1D GP function data, 2D image data, and appropriate context generation needed to replicate experiments from the papers.
- models/ contains implementations of the various models, including training and testing procedures. Some models even offer example scripts for training and visualizing the output.
- utils/ simply contains utility functions used throughout.