uw-forward

• [DONE] SinCosPosEmbed and cls_tkn, clean up
• [DONE] mae decoder finish up
• [DONE] make tensor shapes step through
• [DONE] 1D/2D unfold (called patchify in Kaiming)
• [DONE] loss computation
• [DONE] step through pretraining to verify shapes in GHS
• [DONE] add encode function
• [DONE] early stopping (as TrainMonitor)
• [DONE] walk through GHS fine tuning code (found it confusing!)
• implement custom head
• implement nan aware head loss
• random weight initialization
• pretrained weight initialization
• fine tuning code from GHS
• dataloaders set up for GHS
• train
• comment code for MAE

---

model-bakeoff

• set up the dataset
• train best hyp model inside outer loop
• generate outer loop test predictions with best hyp model
• predict with ensemble on new data; call columns as oi_N
• collate oi outputs
• PLSR optuna pipeline on CHTC
• SPLSR optuna pipeline on CHTC
• Ridge pipeline on CHTC
• Lasso pipeline on CHTC
• ElasticNet pipeline on CHTC
• SVR with Nystroem on CHTC
• (S)PLS-RF on CHTC
• (S)PLS-GPR on CHTC
• (S)PLS-SVR on CHTC
• Stacked (S)PLS on CHTC
• [DONE] Make it model-bakeoff-1
• [DONE] Make hytoptpyt conda env locally, delete older envs
• [DONE] Make .sif container with optuna
• [DONE] Make .sif container with optuna + pytorch

kyle-models

• make toy dataset for CWM
• implement CWM logic
• make sure CSVDataset works for both CWM and regular
• setup pipeline on local machine with hytoptpyt env
• scale on CHTC with hytoptpyt
• pull on nibbler

sophia-lakeview

• make project.py with new dataset
• setup pipeline on local machine with hytoptpyt env
• scale on CHTC with hytoptpyt
• pull on nibbler