Karl Leswing

VAE enable annealing teacher forcing probability during training

11

The VAE doesn't have teacher forcing. The teacher forcing is really needed for larger molecules. Original Code https://github.com/aspuru-guzik-group/chemical_vae/blob/master/chemvae/tgru_k2_gpu.py Moses Code https://github.com/molecularsets/moses/blob/master/moses/vae/model.py#L114-L147

Stale GPG Key in CUCHEM base docker file

1

Fixes issue https://github.com/NVIDIA/cheminformatics/issues/163. When attempting to build the containers you get this issue building cuchem ``` W: GPG error: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64 InRelease: The following signatures couldn't be verified because the public...

When attempting to build the containers you get this issue building cuchem ``` W: GPG error: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64 InRelease: The following signatures couldn't be verified because the public key is not...

https://github.com/rdk/p2rank/issues/59 Stop logging to distribution directory -- instead log to stderr to stderr.

[SUGGESTION] Release Binary Should Log To STDERR instead of to install directory

1

In https://github.com/rdk/p2rank/blob/develop/distro/prank the program logs to a logfile, that is default inside the installation directory. I believe this is the run-script distributed in the release tarball. When installing this on...

You hard-code the checkpoint location to be /training. Putting a .gitignore will mean the user won't have to manually add the folder.

SAScore added to DeepChem

1

This is a P.S. not a real issue. https://github.com/deepchem/deepchem/pull/1182 in case you want to look at the implementation. It uses tf.losses.hingeloss instead of your shifted relu, which doesn't force differentiation...

Create Docker Image

1

Create a Docker Image which opens up the webserver on a port. You can build on top of deepchemio/deepchem

New MoleculeNet Datasets

4

Two years have passed since the publication of MoleculeNet. Since then many strides in supervised learning for molecules have been made. Are all of the datasets from the original paper...

What Can We Do Better?

2

The Moleculenet publication has accomplished much in terms of having standardized problems for supervised learning over chemical structures. However over the past couple of years we have seen some barriers...