Pending Tasks - acappiello/msqc GitHub Wiki

A collection of various ideas to try or things that need to be done. This supplements work being tracked under "Issues."

High Level Theory

• Find the best minimal basis set to use for LL. STO-3G is bad.

Data Analysis

Fits

• Look for a correlation between error and the charge on the atom and the effect of the environment on this:
  • Maybe we can use a linear model to capture any effect.
  • Predict HL as a linear function of LL and charge.
• Correlation across operators:
  • Trends relating KE <--> EN, etc.
  • Look at positive/negative correlations too (don't square data).

Environnments

• Need to get a more concrete was of analyzing environments.
  • Look at the ratio of narrow to diffuse in the HL basis.
• More detailed plotting to show the environment.
  • Look at net charge throughout the box.

Algorithm

• Want to employ more actual machine learning stuff. Bandits and deep learning.
  • https://github.com/rasmusbergpalm/DeepLearnToolbox
• Be more restrictive about when we actually need to recalculate HF (using eps?).

Nuts and Bolts

• Old code (Fragment, etc.) looks kind of scary and could use some cleanup.