Extracting-FSM-From-RNNs/quick_start.py at master · DES-Lab/Extracting-FSM-From-RNNs · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
from aalpy.learning_algs import run_Lstar
from aalpy.oracles import StatePrefixEqOracle
from aalpy.utils import save_automaton_to_file, visualize_automaton

from RNN_SULs import RnnBinarySUL
from TrainAndExtract import train_RNN_on_tomita_grammar, train_and_extract_bp, train_RNN_and_extract_FSM

# learn and extract tomita 3 grammar.
# same can be achieved with train_and_extract_tomita function
rnn = train_RNN_on_tomita_grammar(tomita_grammar=3, acc_stop=1., loss_stop=0.005, train=True)

tomita_alphabet = ["0", "1"]

sul = RnnBinarySUL(rnn)
alphabet = tomita_alphabet

state_eq_oracle = StatePrefixEqOracle(alphabet, sul, walks_per_state=200, walk_len=6)

dfa = run_Lstar(alphabet=alphabet, sul=sul, eq_oracle=state_eq_oracle, automaton_type='dfa',
                cache_and_non_det_check=True)

save_automaton_to_file(dfa, f'RNN_Models/tomita{3}')
visualize_automaton(dfa)

exit()
# train and extract balanced parentheses
bp_model = train_and_extract_bp(path='TrainingDataAndAutomata/balanced()_2.txt', load=False)
print("Print extracted model")
print(bp_model)

# train and learn mealy machine example
coffee_machine_automaton = train_RNN_and_extract_FSM('coffee')
save_automaton_to_file(coffee_machine_automaton, 'CoffeeMachineModel')

mqtt_automaton = train_RNN_and_extract_FSM('mqtt')
save_automaton_to_file(coffee_machine_automaton, 'MqttModel')