Covalent Machine Learning Study

Supplementary Code

This repository contains the complete workflow script (workflow.py) corresponding to this post on AWS Blogs (link to be included upon publication).

The solution here is adapted from this script, originally written by Mateusz Buda. The complete input data can be downloaded here.

Instructions

Before running the workflow, ensure that you have valid AWS credentials and that AWS Batch is correctly configured. The input data should be uploaded to an S3 Bucket as a ZIP file, say data_full.zip.

After the above, proceed with the following:

• Install the required packages (includes Covalent): pip install -r requirements.txt.
• Run the shell command covalent start to start Covalent.

We recommend running the experiment through the argeparse CLI included in workflow.py.

• Run python workflow.py --help to see CLI options that specify the scope of the experiment.

For example, we used the following command to run the experiment in the blog post linked above:

    python workflow.py -B 16 -E 20 -Z 64 128 192 256 -L 0.000075 0.0001 0.000125 -d data_full

Alternatively, just call the workflow function (workflow) normally by passing an arbitrary list of parameters.