supervised-learning-polymers

Evaluation of supervised learning models trained on data from the Open Polymer Prediction dataset. Includes optimization strategies, model evaluation, and performance benchmarking for different supervised learning approaches on polymer datasets.

Reliable models can be deployed for predicting polymer properties, aiding in material design and product development by providing accurate predictions based on supervised learning models trained on polymer datasets.

Documentation

• Current status: PRDs 01, 02, 03, 04, and 13 are accepted. No PRD is active while the structure workbench gets a user-testing round before the next implementation PRD is selected.
• Project backlog
• Benchmark contract
• Chemistry audit
• Geometry groundwork
• Interface discovery GUI backend
• Public interface discovery decision
• Public interface discovery criteria
• Target properties

Quick Start

Smoke Test

For a deterministic check of your setup, you can launch the committed artifact viewer with:

slp-interface-gui tests/fixtures/interface_discovery_run.json --port 8765

Then open http://127.0.0.1:8765 in your web browser.

Structure Viewer and Workbench

Once the smoke test has passed, you can begin inspecting the input data and explore capping strategies by using the structure viewer and workbench.

First, create the necessary local data directories if they do not already exist and add your data files:

mkdir -p data/train data/test

Next, create the necessary local chemistry and geometry artifact directories if they do not already exist:

mkdir -p artifacts/chemistry artifacts/geometry

Then perform a chemistry audit and generate the corresponding geometry artifacts with:

slp-chemistry-audit data/train/train.csv \
  --output-root artifacts \
  --dataset-version open-polymer-train-v1 \
  --chemistry-config-id chemistry-audit-hydrogen \
  --capping-strategy hydrogen

and:

slp-geometry-feasibility artifacts/chemistry/chemistry-audit-hydrogen \
  --output-root artifacts \
  --geometry-config-id geometry-rdkit-hydrogen \
  --input-representation capped_smiles

Then launch the artifact viewer and structure workbench against existing local chemistry and geometry artifacts with:

slp-structure-viewer \
  --chemistry-artifact artifacts/chemistry/chemistry-audit-hydrogen \
  --geometry-artifact artifacts/geometry/geometry-rdkit-hydrogen \
  --port 8765

Then open http://127.0.0.1:8765 in your web browser.

Results

Exploratory data analysis of the dataset can be found in eda.ipynb.

Evaluation of the capping strategy can be found in capping-strategy-audit.md, which provides a detailed comparison of uncapped, hydrogen-capped, and carbon-capped chemistry inputs and their impact on geometry coverage and failure modes.

License

MIT — see LICENSE.