SACRO-ML
SACRO-ML copied to clipboard
AI-SDC
Collection of tools and resources for managing the statistical disclosure control of trained machine learning models
updates: - [github.com/psf/black: 24.4.0 → 24.4.2](https://github.com/psf/black/compare/24.4.0...24.4.2)
![pre-commit-ci[bot] avatar](https://avatars.githubusercontent.com/in/68672?v=4 "pre-commit-ci[bot] avatar"){kind=avatar}
Some of the code has been updated at various points where the docstrings and type hints haven't been kept in synch. For example, the function below declares that it returns...
The test code coverage dropped slightly because of the addition of the new datasets, which was merged in before the code coverage reporting was fully fixed so it just needs...
Hi all, perhaps especially @yolaj-nhs if you have capacity. I had an ... interesting discussion with one of our TREs yesterday who said the first they had known about the...
Maybe even phase out safekeras since tensorflow privacy is so dependent on out of data packages
Structural attack fails in unnecessary disclosure. with xgboost if min_child_weight is not specified
line 126/128, error is comparison between NoneType and float for optional parameters. Fix is to test whether param is none first and take appropriate action. Will be true for max_deth...
we currently report the ratio of the proportion of training records that. are vulnerable to the proportion of test records that are vulnerable if we're to. take the logarithm of...
How to make the data splits of the LIRA reproducible? I noticed in the code it used random to generate indices to select rows of data, see line 303 these_idx...
- but should make sure it warns user that what they are doing is risk and usually unnecessarily so - warning should tell user how to apply constraints if they...
Basic idea: dataset D split into training set Tr and test set Te 1. concatenate train and test sets (so the top |Tr| records come from the training set and...