Installing into a conda virtual environment

[breisfeld]

For those people interested in installing the package into a conda virtual environment, I was successful using the following:

$ conda create -n guacamol -c conda-forge rdkit python=3.7
$ conda activate guacamol

The requirements.txt file in the repo seems incorrect.

I changed git+https://github.com/molecularsets/moses.git@master#egg=moses to
git+https://github.com/molecularsets/moses.git@master#egg=molsets

Also, the version of numpy needs to be updated because of the tensorflow requirements. The final requirements.txt file looked like

guacamol==0.5.3
matplotlib==3.0.2
torch>=1.0.0
joblib==0.12.5
numpy==1.18
tqdm==4.26.0
cython==0.29
nltk==3.4.5
flake8==3.5.0
git+https://github.com/molecularsets/moses.git@master#egg=molsets

After installing git-lfs, I then installed the other required packages:

$ pip install -r requirements.txt

[JoshuaMeyers]

Hey Brad, thanks for the information and for pointing out the inconsistency in requirements.txt. The above solution will work when using some of the models in the baselines repo, but to use the models from the moses suite I believe you will need to revert to an old version of the moses code (the package has been renamed to molsets more recently).

Unfortunately the GuacaMol baselines predate the availability of Moses on PyPi but a preliminary search suggests that using git+https://github.com/molecularsets/moses.git@96c77d52c688dcac8e2962d78965175e7824c48c#egg=moses in requirements.txt should work.

As you also point out, there are a number of dependencies whose versions have drifted, particularly numpy and keras (required by FCD). We will aim to submit a PR for these changes soon.

[breisfeld]

Hi Joshua, I appreciate the comments. I'll give that Moses branch a try and stay tuned for updates to the GuacaMol repo.

[JoshuaMeyers]

Hey @breisfeld, the installation issues should be resolved by #14. Thanks for raising