Metrics for batch adjustment effectiveness #5
Description
- Accuracy of these (Discriminators between batches #3)
- How much has actually changed? (MSE between before & after)
- With MNIST, are we still able to accurately predict the true signal (i.e. which digit it is)? Predict MNIST digit #26
- Other (what do other papers use? What do I mention in my prospectus?)
These metrics should ideally be automatically (or optionally?) output when running Confounded.