From 641c4dd98c97cb628f31931458a9fc5563e17e81 Mon Sep 17 00:00:00 2001 From: Aadyaa Maddi Date: Fri, 13 Nov 2020 12:20:30 +0800 Subject: [PATCH] Add comment to explain support for LR --- .../membership_inference_attack.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/tensorflow_privacy/privacy/membership_inference_attack/membership_inference_attack.py b/tensorflow_privacy/privacy/membership_inference_attack/membership_inference_attack.py index f5a0f6a..c60535f 100644 --- a/tensorflow_privacy/privacy/membership_inference_attack/membership_inference_attack.py +++ b/tensorflow_privacy/privacy/membership_inference_attack/membership_inference_attack.py @@ -188,6 +188,10 @@ def run_seq2seq_attack(attack_input: Seq2SeqAttackInputData, the attack result. """ attack_input.validate() + + # The attacker uses the average rank (a single number) of a seq2seq dataset + # record to determine membership. Hence, only Logistic Regression is supported, + # as it makes the most sense for single-number features. attacker = models.LogisticRegressionAttacker() prepared_attacker_data = models.create_seq2seq_attacker_data(