Examine mislabelled data

AI Discussions
1

Any strategies on how to manually identify mislabelled data?

I’m only doing a binary classification task with sources from two pools of 1-dimensional data.

Please leave your suggestions!

Thanks
Yuhan Chiang

2

Find the index values there the prediction and the label are incorrect.
Inspect the examples that have those indices.

1 Like
3

I think in the dev (or validation set) in keras randomly shuffles the data every iteration, therefore I might only be able to do it on the test set. Is it customary to use it only on the train set?

Also, are there any functions that I could use to manually do this? I was using the model.evaluate function and it cannot map out all the examples. I arrange my examples into numpy arrays.

Thank you for your reply, I hope you can just give me a little more hints.

Yuhan