Skip to content

Latest commit

 

History

History
30 lines (21 loc) · 2.78 KB

README.md

File metadata and controls

30 lines (21 loc) · 2.78 KB
Raw

deployment

ChEmbVis - Chemical Embedding Visualization

This project is a (currently developed) application dedicated to the interactive analysis and exploration of chemical embeddings and projections. A live demo is available at https://chembvis.pueh.xyz/.

Several submodules make up the application:

  • api: Main API exposing projections, MSO, rdkit, ...
  • client: Interactive web application
  • api_umap: API for pretrained parametric UMAP
  • api_tsne: API for pretrained parametric TSNE
  • notebook: Jupyter notebook with several experiments

One reason why the API is split into several submodules is because they depend on different tensorflow versions, as the main API requires tensorflow==1.15.0, while the others depend on tensorflow>=2.0.

Getting Started

The easiest way to run the whole application is by using docker[-compose]. To start local development, simply use docker-compose start. Please note that building the images might take a while, as we are using large dependencies such as RDKit and Tensorflow.

When the building is done, the frontend is available at http://localhost:3000 and the backend at http://localhost:5000, with a Swagger API documentation available at http://localhost:5000/api/spec.

Please note that this project is actively developed such that no optimized builds/deployments are available yet.

_shared Data

Due to the size of the parametric models, the contents of the _shared folder are not provided in this repository. Each package has a _shared folder which contains such data. For easy setup, please download the entire shared data and extract it into the repository: https://drive.google.com/file/d/1F-QFxKd8A1rnXmyOAsfixF_CCUQYtmwY/view?usp=sharing Otherwise, take a look at the individual repositories for further details.

GPU Support:

By default, the GPU enabled tensorflow-gpu is installed and docker-compose uses GPU capabilities for some containers (requires docker-compose 1.28+). For WSL2 users, please see https://docs.nvidia.com/cuda/wsl-user-guide/index.html first. For docker and docker-compose related issues, see https://docs.docker.com/config/containers/resource_constraints/#gpu and https://docs.docker.com/compose/gpu-support/#enabling-gpu-access-to-service-containers. If you want to use pure CPU implementations, please replace all tensorflow-gpu occurances with tensorflow in the corresponding environment.yml files. Also, you need to remove the deploy entries in the docker-compose.yml.