wangleiphy/ml4p

Machine learning for physicists

ml4p's Issues

[A project suggestion] ML with data: Molecular Properties prediction

This issue suggests an open competition that took place in Kaggle four years ago.

Problem Page

[TL;DR]

Scientific ML with data: Use the data to develop an algorithm to predict the interaction between two atoms in a molecule.

Background in DFT, MD would be helpful.

Pros

• It requires not only brute force ML practice, but also domain knowledge in Computational Physics/Chemistry.
• It is a quantum many-body problem that all Physics students would be familiar with.
• Well-documented and explained notebooks are publicly available, with active updates recently after four years.
• Kaggle provides a well-established ranking and evaluation system.

Cons

• With so many open-source notebooks, it is hard to avoid plagiarism.
• Most codes are written with TensorFlow Keras instead of Jax.
• The computational demand for a good score is high.

Find your place in the score board! 😃