thunlp/OpenBackdoor

Questions for clean- and poison-data in poison_data directory

Closed this issue · 1 comments

WencongY commented

Hello again,

I have a few questions about the data under the ./poison_data/ directory and hope you could kindly assist:

  1. Would you please help me understand the columns in files such as train-clean.csv, test-poison.csv? The column names in those files are marked as 0, 1 and I wasn't able to find what they represent, some clarification would be appreciated.

  2. I'm trying to use my own poison data by uploading my train-/dev-/test-clean, and train-/dev-/test-poison data, and by setting load: true in the config file. However, I'm a bit confused by how the test-eval.csv and test-detect.csv were created in the 'eval' and 'detect' modes. would you please explain?

  3. Additionally, I thought the test-poison.csv file contained only poison test data, but then the test-poison.csv file seems to change and becomes larger during the experiment, even if I set load:true. What changes may have been made to this file in the process?

Thank you very much for your timely response!

Best,
Wencong

WencongY commented

Hello,

I've figured it out, lol...
Closing the ticket now.

Best,
Wencong