10/4 Meeting Notes

[panalluri]

• See how well model performs on training datasets --> shows if NN actually learned properly

• MSE b/n targets and predictions is useful

• Figure out what kernel size means?

• Shuffle dataset before splitting into training and testing (only relevant for big datasets) --> we want to give snapshots scattered throughout the system's evolution

• Sensitivity analyses --> (1) how many different time points are necessary for the training dataset, how many till predictions on testing dataset are solid? run experiments; (2) how many different regional timepoints of n snapshots are necessary for the training dataset, how many till predictions on testing dataset are solid? run experiments

• Run NN on training dataset --> compare predictions and targets here!

• Qualitatively look at the trends in my dataset (e.g. Q_star target, other input vars) --> which regions have which characteristics?

• Laure can train on data of some dims, then test on completely different dims, and this works --> figure out why and implement this

• IDP: make specific list of research goals for next meeting

• --> look at Ryan's 6 month check-ins for structure and examples

• Montse meeting: do I need a committee soon? What do I need ready for the orals? Who grades me on that?