Closed tommylau-exe closed 4 years ago
For posterity, the current F1 Score for each category (as calculated by the recently pushed code) is roughly as follows:
Seems to correlate pretty well with representation in the data set.
In addition, the confusion matrix generated by the recently pushed code looks roughly like the following:
Predicted Low-Calorie | Predicted Average-Calorie | Predicted High-Calorie | |
---|---|---|---|
Actual Low-Calorie | 10765 | 2437 | 52 |
Actual Average-Calorie | 2065 | 32897 | 979 |
Actual High-Calorie | 74 | 2100 | 2977 |
This PR contains code that was used to calculate how balanced the calorie labels (classes) are in the data set with our current thresholds. The results are as follows: