Skip to content

Latest commit

 

History

History
73 lines (44 loc) · 1.48 KB

d83e-dde5.md

File metadata and controls

73 lines (44 loc) · 1.48 KB
1.4 Satisficing and optimizing metrics

Example:

1. Accuracy 
2. Running time

cost = accuracy - O.5 * runningTime

maximize accuracy, running time <= 100ms

Optimize Accuracy, Satisficing Running time

**If you have N metrics, 1 to optimize, N - 1 to satisficing **

Another example: Wakewords / Trigger words to wake up voice device

Accuracy # of fake positive

Maximize accuracy while at most 1 false positive every 24 hours

1.5 Train/dev/test distribution

Example 1: Cat classification dev/test set

Regions:

|dev set|test set| |--|--| |US, UK|China, India, Australia|

Example 2: loan estimation

use zip code to predict loan they use middle income zip codes as dev set use low income zip code as test set the result is disappointting

Conclusion: Choose a dev set and test set to reflect data you expect to get in future and consider important to do well on. Set the dev/test set from same distribution.

1.6 Size of dev and test sets
1.6.1 Percentage of each set

Old Era: 70% Train 30% Test

60% Train 20% Dev 20% Test

Now: Given you have 1 million data set 98% Train 1% Dev 1% Test So in the modern times, we have more data, so have less than 20% for test/dev set is more reasonable.

1.6.2 Size of test set

Set you test set to be big enough to give high confidence in the overall performance of your system

If you do not need high confidence in performance, you can only have 2 sets: dev and train. no test set.