Retrain with New Labels

$\frac{0\mathrm{.}0}{3\mathrm{.}0}$ points $($graded$)$

Now suppose that instead we want to retrain our classifier with the new labels. In other words, rather than

training the model to predict the original digits and then taking those predictions modulo $3,$ we explicitly train

the model to predict the digits modulo $3$ from the original image.

How do you expect the performance to change using the new labels?

Increase

Decrease

Stay the same

Implement

in

main.py to perform this new training; report the new error rate.

Error rate when trained on labels mod $3$: