googleinterns / amaranth

Apache License 2.0
2 stars 0 forks source link

Change high calorie threshold to 300 #16

Closed tommylau-exe closed 4 years ago

tommylau-exe commented 4 years ago

This change adjusts how data in the FDC data set is classified as high calorie. Now, any dish with over 300kcals per 100g of food is "high calorie." This better balances the classes in the data set, and is a more realistic number based on people eating 1.5-2kg of food per day.

tommylau-exe commented 4 years ago

Putting the new rough metrics in the PR here for posterity:

Data Set Class Balance

ML Model Metrics

Confusion Matrix

Predicted Low-Calorie Predicted Average-Calorie Predicted High-Calorie
Actual Low-Calorie 11203 1712 591
Actual Average Calorie 1602 13103 2120
Actual High-Calorie 752 2472 20791