Why compute the accuracy and roc at every batch instead of once at the end?

DisaitekAI / membership_inference_attack

Implementation for the experiments in the blogpost "Demystifying the Membership Inference Attack" at https://www.mlsecurity.ai/post/demystifying-the-membership-inference-attack

MIT License

1 stars 0 forks source link

Why compute the accuracy and roc at every batch instead of once at the end? #15

Closed rodgzilla closed 5 years ago

rodgzilla commented 5 years ago

https://github.com/DisaitekAI/membership_inference_attack/blob/de3ab672d0850a53badcd7bfb5f5206b86dd9d64/src/utils/statistics.py#L44-L45

EtoiledeScauchy commented 5 years ago

The objective is to show the evolution of the performance with these indicators during the training