Logistical Concerns for Deep Learning

Overview

This week we talked about non-deep learning tasks that are necessary for effective modeling.

• Data cleaning and storage
  • Pandas, Polars, and Databases
• Model evaluation and test set construction
  • Structure in the data and information leakage
• Case Study: Workflow to predict viral infection potential
  • Separating data prep from modeling
  • Using cross validation folds stratified with respect to species
  • How do we think about predicting cases that were not observed?
  • Example pipeline
• Tracking Hyperparameters and Experiments
  • General workflow
  • KerasTuner, Ray, and Ax
  • Example in Ax

Meeting Notes

• The slides from today can be found here.