DeepDFA

Official repository for the paper "DeepDFA: Automata Learning through Neural Probabilistic Relaxations" accepted by the 27th European Conference on Artificial Intelligence (ECAI 2024).

Requirements

• Pythomata
• Pytorch

How to execute the code

To reproduce the experiments in the paper execute the file do_experiments.py. Run do_experiments.py --help for a complete list of the parameters used by the script (below).

USAGE: do_experiments.py [flags]
flags:

do_experiments.py:
  --dataset_dir: path to the datasets
    (default: 'datasets_Tomita/')
  --error_rate: portion of errated labels in the training dataset
    (default: '0.0')
    (a number)
  --experiment: which experiment to execute, choose one in the list ['tomita', 'random_dfa']
    (default: 'tomita')
  --ground_truth_dfa_dir: path to save ground truth automatons
    (default: 'DFA_ground_truth/')
  --plot_dir: path to save the plots
    (default: 'Plots/')
  --predicted_dfa_dir: path to save predicted automatons
    (default: 'DFA_predicted_nesy/')
  --results_dir: path to save the results
    (default: 'Results/')

Examples:

To learn the Tomita languages from data execute the script with default parameters

python do_experiments.py

To learn the random DFAs in the directory random_DFAs run

 python do_experiments.py --dataset_dir datasets_random_dfa/ --experiment random_dfa

To repeat the expermints on a corrupted version of the datasets, with 1% of flipped labels, add to the commands above --error_rate 0.01