Experiments on realistic conditions
This contains the code we used for our paper, "On Realistically Attacking Tor with Website Fingerprinting."
We wrote "notes" for each of the following, which describes how to run and use the files:
Training update. For testing a training set updating scheme.
Splitting. For split finding, split decision, and pre-splitting. About 100 MB.
Classifier. Takes packet sequences collected in the wild as input, and performs time-based splitting, classification-based splitting, and kNN classification on them. About 700 MB.
Tor Logging. Implementation of Tor logging for the above.
This is necessary for the file format required by the classifier.
Note that the classifier is only allowed to look at time and direction;
other information is not available to the website fingerprinting attacker.