cnp

Conditional Neural Processes

Tensorflow 2 generic implementation of the conditional neural processes (CNP).

Usage

To train the model on a regressive task or on MNIST

python train.py -t mnist
python train.py -r regression

On MNIST, the CNP is trained with a context size varying between 10 and 100 points.

On the regression task, the CNP is trained with a context size varying between 2 and 10 points.

To see all training options use :

python train.py -h

To see visualizations use

tensorboard --logdir ./logs