neuralprocesses
neuralprocesses copied to clipboard
A framework for composing Neural Processes in Python
I've been converting my code from tensorflow to pytorch and it's much easier to get it training faster. However, the performance after n epochs is worse in torch. After lots...
Hey there. I've been revisiting neural processes for use in a project dealing with simple time series data (sets of `x, y` values for which we'd like to make future...
I have a small time series dataset of size 500x10 (500 time steps, 10 features). I want to make predictions several time steps into the future conditioned on the first...
[FR] Ability to draw cheap AR samples with subset-condition-predict procedure with `nps.ar_predict`
In our [AR CNP paper in ICLR](https://openreview.net/forum?id=OAsXFPBfTBh), we describe some ways to make AR sampling cheaper through a subset-AR sample-condition-predict procedure. For example, in Appendix K: > The AR samples...
Currently, calling `B.jit` on a `logpdf`-based objective works for a model with a single context set and single target set. However, a `ValueError` is raised in the case where a...
[The Antarctica data generator always loads PyTorch](https://github.com/wesselb/neuralprocesses/blob/7aeb4f6fbb87c32ec75f86af2d516d03a970e5d7/neuralprocesses/data/antarctica.py#L2), which forces scripts that intend to use only TF to also install Torch or [even causes these scripts to crash]. The data generator...
Currently, with a TensorFlow backend, I believe all the array inputs to `nps.loglik`, `nps.ar_loglik` and `nps.ar_sample` need to be TensorFlow tensors, e.g. `tf.EagerTensor`s. It would be convenient if users with...