TabICLmolprop Demo (CPU)

This Space compares TabICL, RandomForest, and CatBoost on molecular regression.

Featurizer options: RDKit2D, CheMeleon, or Mordred
Default data: fixed DCN split with 100 train and 10 test rows
Custom data schema: CSV with exactly two columns in this order: smiles,value

Full Benchmark Repo: https://git.rwth-aachen.de/avt-svt/public/tabpfn-molprop

Featurizer

Use default DCN 100/10 split

Train CSV (smiles,value)

Test CSV (smiles,value)

Metrics

TabICL parity

RandomForest parity

CatBoost parity

Status

Acknowledgements

This code uses CheMeleon. The code also uses the TabICLv2 Model. Example dataset from here: Graph neural networks for ignition quality prediction