This is a PyTorch implementation of various Neural Process (NPs) variants, including Standard NPs [1], Self-attentive NPs [2], and NPs with Bayesian Aggregation [3].
Plots taken from [3].
[1] Garnelo et al., "Neural Processes", ICML 2018 Workshop on Theoretical Foundations and Applications of Deep Generative Models
[2] Kim et al., "Attentive Neural Processes", ICLR 2019
[3] Volpp et al., "Bayesian Context Aggregation for Neural Processes", ICLR 2021, cf. https://github.com/boschresearch/bayesian-context-aggregation
First install metalearning_benchmarks from here.
Then clone this repository and run
pip install .
from the source directory.
To get familiar with the code, have a look at the example script ./scripts/run_neural_process.py.
This code is still in development and thus not all features are thoroughly tested. Some features may change in the future. It was tested only with the packages listed in ./setup.cfg.
This code is licensed under the AGPL-3.0 license and is free to use by anyone without any restrictions.