cracking-rnn

A recurrent neural network for password cracking.

Separate dataset into train/valid/test:

$ python preprocess.py

Note that preprocess.py does not shuffle its input. You should shuffle your dataset before feeding it to preprocess.py

Train your own model:

$ python train.py

Sample all passwords with possibility above threshold:

$ python sample.py [--threshold <threshold>]

Evaluate how well your model performed:

$ python eval.py <results_file> <test_data_file>

Estimating Guess Numbers with Monte Carlo

Sampling a large number of passwords may take a very long time. Therefore, we use Monte Carlo methods to estimate the guess number of each password in the test set.

Guess number: number of guesses needed to crack a particular password.

The code for Monte Carlo requires two input files: one that contains k possibilities randomly sampled from the entire distribution (e.g. k = 10000) and another that contains the possibility of each password in the test set.

Randomly sample k possibilities from the distribution:

$ python sample-for-monte-carlo.py [--sample_size <sample_size>]

Assign possibilities to each password in the test set:

$ python assign-probs.py

Evaluate how well your model performed:

$ python eval-monte-carlo.py

Note that due to copyright reasons, the code for Monte Carlo is not included in this repository.