harvard-edge / dataperf-speech-example

Example workflow for our data-centric speech benchmark
17 stars 11 forks source link

Balanced scoring and separate selection files #9

Closed colbybanbury closed 2 years ago

colbybanbury commented 2 years ago

New Baseline Scores:

* This poor accuracy is because I didn't update the CleanLabs method so it returns a very unbalanced training set that leads the model to predict unknown most of the time