instance_attributions_NLP icon indicating copy to clipboard operation
instance_attributions_NLP copied to clipboard
verified publisher icon successar

RUNTIME ERROR: undefined external variable: BERT_MODEL_NAME

Open Srijha09 opened this issue 3 years ago • 7 comments

I am getting a runtime error while training. image

Can you please help me solve this error?

Srijha09 avatar Oct 27 '22 19:10 Srijha09

Hi,

Can you include BERT_MODEL_NAME="bert-base-uncased" as env var to the command also? If it still gives an error, let me know.

successar avatar Nov 01 '22 22:11 successar

Thank you, the code works.

Also, does it take a really long time to run the influence value command using bash? The code almost ran for 4 hours.

Srijha09 avatar Nov 04 '22 16:11 Srijha09

Yes, I believe the most time consuming part is calculating influence via this https://github.com/successar/instance_attributions_NLP/blob/master/influence_info/influencers/influence_functions.py since it involves calculating hessians vector products.

successar avatar Nov 04 '22 18:11 successar

Thank you, also just a quick doubt, for the nearest neighbors, once you get the influence_values.npy file, is there a threshold value that is used to remove 500 training data points in order to retrain it?

Srijha09 avatar Nov 04 '22 20:11 Srijha09

No, we just remove the top 500 examples before retraining.

successar avatar Nov 11 '22 23:11 successar

Thanks, just to confirm, are these top 500 samples randomly selected or is there an inherent ranking based on some criteria?

Srijha09 avatar Nov 14 '22 14:11 Srijha09

It depends on the value of the influence method. So in case of NN, you would remove the top-500 examples with highest NN similarity for a given test point. The value of the influence method is the criteria.

successar avatar Nov 17 '22 04:11 successar