The best meal I found was AGORTV. In general, ORV + optional P have good synergy, and must be included, and it’s also good to sample a few desserts from AGT. These were the top twenty meals:
I trained 100 small models with a bit of weight regularization and averaged their predictions together. I think a better approach would be to train on diffs between meals, if anyone wants to try that out.
Predictions:
The best meal I found was AGORTV. In general, ORV + optional P have good synergy, and must be included, and it’s also good to sample a few desserts from AGT. These were the top twenty meals:
```
AGORTV
GOPRTV
GOPRV
GORTV
ABOPRV
AGORV
AOPRTV
AGOPRTV
GORV
AOPRV
ADGORTV
ABOPRTV
GMOPTV
GPRTV
ABPRTV
AGOPRV
AHORTV
BGOPTV
DGOPRTV
ADOPRTV
```
Method:
I trained 100 small models with a bit of weight regularization and averaged their predictions together. I think a better approach would be to train on diffs between meals, if anyone wants to try that out.