Cornell Tech cs5785 Final Project
- implement Bert Model
- analyze using techniques discussed in lecture
- follow rubric
Given a toolbox of ML techniques, how do we analyze their performance and decide what models to apply and when?
12/2:
- error analysis
- bias variance analysis
12/4:
- Confusion matrix
- Sensitivity (true positive rate) and Specificity (true negative rate) include these
- Balanced accuracy (average of tpr and tnr)
- Precision, Recall, and F-score
- ROC
- Learning curves
- show performance as a function of training set size
- not realistic for NN
- Loss Curve
- Loss/epoch -> necessary for NN