replication.md

DISAE Experiments Replication

These are the steps in replicating DISAE Experiments using DGX-Lei workstation or the Hunter DLS Server. Please follow these steps after cloning the repository and downloading the data. Instructions in doing so are in the README.md file, Steps 2 and 3.

----------- INSTRUCTION -----------

1. Run a Docker Containter

Example:

nvidia-docker run --name pytorch -it --network=host --rm -v /raid/home/username/:/username nvcr.io/nvidia/pytorch:19.07-py3

2. Create an RDKit environment

conda create -n rdkit-env -c rdkit rdkit libboost=1.65.1
conda activate rdkit-env # If this line does not work, use 'source activate rdkit-env' instead 

3. Install packages

conda install -c conda-forge dataclasses scipy networkx tensorflow scikit-learn

4. Install version 2.3.0 transformers using pip

pip install transformers==2.3.0

5. (Optional) Install pytorch and torchvision according to CUDA Version of GPU

nvcc --version # Check CUDA Version
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch # Replace 10.1 with appropriate version

6. View README.md and follow Step 4