jleechung / neural-processes

PyTorch implementation of latent neural processes.

Neural processes

PyTorch implementation of (latent) neural processes for function regression.

See src/model for components of the NP, and here for a walkthrough of the model and training.

Set up an environment with:

conda env create -n neural-processes -f environment.yaml

Examples

Visualise the predictive mean and variance from the trained NP given increasing number of context points:

References

See the official Tensorflow implementation.