Kaggle Predicting Molecular Properties

My solution in this Kaggle competition "Predicting Molecular Properties", 19th place.

Acknowledgement

I used xyz2mol to parse molecular structure. source/01_preprocess/xyz2mol.py is forked from below repository.

• xyz2mol
  https://github.com/jensengroup/xyz2mol

I employed jo's tips to handle xyz2mol in source/01_preprocess/xyz2mol_jo.py

• Using RDKit for Atomic Feature and Visualization
  https://www.kaggle.com/sunhwan/using-rdkit-for-atomic-feature-and-visualization

I used train_ob_charges.csv and test_ob_charges.csv which are output of Alexandre's notebook. Please put them in the input directory when you run my code.

• V7 Estimation of Mulliken Charges with Open Babel
  https://www.kaggle.com/asauve/v7-estimation-of-mulliken-charges-with-open-babel

Even though, my solution doesn't depend on chainer-chemistry directly, my implementations are inspired by it.

• Chainer Chemistry: A Library for Deep Learning in Biology and Chemistry
  https://github.com/pfnet-research/chainer-chemistry

References

• Neural Message Passing with Edge Updates for Predicting Properties of Molecules and Materials
  Peter Bjørn Jørgensen, Karsten Wedel Jacobsen, Mikkel N. Schmidt
  https://arxiv.org/abs/1806.03146
• Weisfeiler and Leman Go Neural: Higher-order Graph Neural Networks
  Christopher Morris, Martin Ritzert, Matthias Fey, William L. Hamilton, Jan Eric Lenssen, Gaurav Rattan, Martin Grohe
  https://arxiv.org/abs/1810.02244