Multitask

Optimizing reactions using multitask Bayesian optimization. Based on the paper Accelerated Chemical Reaction Optimization Using Multi-Task Learning.

Setup

Clone the project:

git clone https://github.com/sustainable-processes/multitask.git

Install poetry and then run the following command to install dependencies:

poetry install

Additionally, until this issue is fixed, you will need to manually install the ord-schema dependencies:

poetry run pip install -r ord-requirements.txt

You can run commands from inside the poetry virtual environment in one of two ways:

• Option 1: Put poetry run in front of every command (e.g., poetry run dvc repro) as we did above to install extra requirements.
• Option 2: Activate the virtual environement and run everything as normal:
  • Run poetry show -v. Copy the path specified after "Using virtualenv"
  • Activate the virtual environemnt at that path. So, if the path is /Users/Kobi/Library/Caches/pypoetry/virtualenvs/multitask-z7ErTcQa-py3.7, you would run source /Users/Kobi/Library/Caches/pypoetry/virtualenvs/multitask-z7ErTcQa-py3.7/bin/activate

Apple M1

You might run into some issues when installing scientific python packages such as Summit on Apple M1. Follow the steps below to install via pip:

arch -arm64 brew install llvm@11 
brew install hdf5
HDF5_DIR=/opt/homebrew/opt/hdf5 PIP_NO_BINARY="h5py" LLVM_CONFIG="/opt/homebrew/Cellar/llvm@11/11.1.0_3/bin/llvm-config" arch -arm64 poetry install

Replace the llvm path with the version of llvm installed by brew.

CLI

There is a multitask CLI available. You can see all the commands by running multitask --help

Regnerating the results

1. ETL: Transform data from spreadsheets into ORD protobuf files.

   multitask etl all

2. Benchmark training and Benchmarking using Lightning:

   lightning run app  --name multitask --open-ui False --without-server lightning_study.py

   Some extra flags to add:

   • --env WANDB_API_KEY=<YOUR_API_KEY>: Add your weights and biases API key
   • --cloud: Run on lightning cloud instead of locally
   • --env lightning_cloud=True: Tag these runs as running in the cloud

3. Generate plots

   multitask viz all-suzuki --include-tags lightning_cloud --wandb-entity "ceb-sre" 

Development

Please format your code using black. Jupyter notebooks, which are for exploration, don't need to be formatted.

Directory Structure

├── data/  # Store all data and results here (tracked mainly by DVC)
├── figures/  # Store figures in this directory and track using dvc
├── multitask # python package with key functionality
├── nbs # Exploratory Jupyter notebooks
├── poetry.lock  # Poetry lock file
└── pyproject.toml # Poetry configuration file

• Put final code for pipelines in multitask, and add the code to pipelines as described in the previous section.
• Typer is great for turning a python function into a command line script. It's already installed, so you might as well use it.

Citing

If you find this project useful, we encourage you to

• Star this repository ⭐
• Cite our paper.

@article{Taylor2023,
  doi = {10.1021/acscentsci.3c00050},
  url = {https://doi.org/10.1021/acscentsci.3c00050},
  year = {2023},
  month = apr,
  publisher = {American Chemical Society ({ACS})},
  volume = {9},
  number = {5},
  pages = {957--968},
  author = {Connor J. Taylor and Kobi C. Felton and Daniel Wigh and Mohammed I. Jeraal and Rachel Grainger and Gianni Chessari and Christopher N. Johnson and Alexei A. Lapkin},
  title = {Accelerated Chemical Reaction Optimization Using Multi-Task Learning},
  journal = {{ACS} Central Science}
}