X-Git-Url: https://git.njae.me.uk/?a=blobdiff_plain;f=cipherbreak.py;h=342f5ed1fe6f6f593313e86751721aebe724adc3;hb=7bdef48d00d4f7fd2e9e1f14d6438d489adc54d6;hp=a2eecad9cb34f3bb596e2e38e22ed60c1ccca840;hpb=7c5fd4061335669401bb298b5cec519b1f9afbc8;p=cipher-tools.git diff --git a/cipherbreak.py b/cipherbreak.py index a2eecad..342f5ed 100644 --- a/cipherbreak.py +++ b/cipherbreak.py @@ -40,6 +40,18 @@ from language_models import * # timeit.timeit('keyword_break(c5a)', setup='gc.enable() ; from __main__ import c5a ; from cipher import keyword_break', number=1) # timeit.repeat('keyword_break_mp(c5a, chunksize=500)', setup='gc.enable() ; from __main__ import c5a ; from cipher import keyword_break_mp', repeat=5, number=1) + +def index_of_coincidence(text): + stext = sanitise(text) + counts = collections.Counter(stext) + denom = len(stext) * (len(text) - 1) / 26 + return ( + sum(max(counts[l] * counts[l] - 1, 0) for l in string.ascii_lowercase) + / + denom + ) + + transpositions = collections.defaultdict(list) for word in keywords: transpositions[transpositions_of(word)] += [word] @@ -209,35 +221,110 @@ def keyword_break_worker(message, keyword, wrap_alphabet, fitness): wrap_alphabet, fit, sanitise(plaintext)[:50])) return (keyword, wrap_alphabet), fit -def monoalphabetic_break_hillclimbing(message, max_iterations=10000000, - alphabet=None, fitness=Pletters): - ciphertext = unaccent(message).lower() - if not alphabet: - alphabet = list(string.ascii_lowercase) - random.shuffle(alphabet) - alphabet = cat(alphabet) - return monoalphabetic_break_hillclimbing_worker(ciphertext, alphabet, - max_iterations, fitness) - -def monoalphabetic_break_hillclimbing_mp(message, workers=10, - max_iterations = 10000000, alphabet=None, fitness=Pletters, chunksize=1): +# def monoalphabetic_break_hillclimbing(message, max_iterations=10000000, +# alphabet=None, fitness=Pletters): +# ciphertext = unaccent(message).lower() +# if not alphabet: +# alphabet = list(string.ascii_lowercase) +# random.shuffle(alphabet) +# alphabet = cat(alphabet) +# return monoalphabetic_break_hillclimbing_worker(ciphertext, alphabet, +# max_iterations, fitness) + +# def monoalphabetic_break_hillclimbing_mp(message, workers=10, +# max_iterations = 10000000, alphabet=None, fitness=Pletters, chunksize=1): +# worker_args = [] +# ciphertext = unaccent(message).lower() +# for i in range(workers): +# if alphabet: +# this_alphabet = alphabet +# else: +# this_alphabet = list(string.ascii_lowercase) +# random.shuffle(this_alphabet) +# this_alphabet = cat(this_alphabet) +# worker_args.append((ciphertext, this_alphabet, max_iterations, fitness)) +# with Pool() as pool: +# breaks = pool.starmap(monoalphabetic_break_hillclimbing_worker, +# worker_args, chunksize) +# return max(breaks, key=lambda k: k[1]) + +# def monoalphabetic_break_hillclimbing_worker(message, alphabet, +# max_iterations, fitness): +# def swap(letters, i, j): +# if i > j: +# i, j = j, i +# if i == j: +# return letters +# else: +# return (letters[:i] + letters[j] + letters[i+1:j] + letters[i] + +# letters[j+1:]) +# best_alphabet = alphabet +# best_fitness = float('-inf') +# for i in range(max_iterations): +# alphabet = swap(best_alphabet, random.randrange(26), random.randrange(26)) +# cipher_translation = ''.maketrans(string.ascii_lowercase, alphabet) +# plaintext = message.translate(cipher_translation) +# if fitness(plaintext) > best_fitness: +# best_fitness = fitness(plaintext) +# best_alphabet = alphabet +# print(i, best_alphabet, best_fitness, plaintext[:50]) +# return best_alphabet, best_fitness + + +def monoalphabetic_break_hillclimbing(message, + max_iterations=20000, + plain_alphabet=None, + cipher_alphabet=None, + fitness=Pletters, chunksize=1): + return simulated_annealing_break(message, + workers=1, + initial_temperature=0, + max_iterations=max_iterations, + plain_alphabet=plain_alphabet, + cipher_alphabet=cipher_alphabet, + fitness=fitness, chunksize=chunksize) + + +def monoalphabetic_break_hillclimbing_mp(message, + workers=10, + max_iterations=20000, + plain_alphabet=None, + cipher_alphabet=None, + fitness=Pletters, chunksize=1): + return simulated_annealing_break(message, + workers=workers, + initial_temperature=0, + max_iterations=max_iterations, + plain_alphabet=plain_alphabet, + cipher_alphabet=cipher_alphabet, + fitness=fitness, chunksize=chunksize) + + +def simulated_annealing_break(message, workers=10, + initial_temperature=200, + max_iterations=20000, + plain_alphabet=None, + cipher_alphabet=None, + fitness=Pletters, chunksize=1): worker_args = [] - ciphertext = unaccent(message).lower() + ciphertext = sanitise(message) for i in range(workers): - if alphabet: - this_alphabet = alphabet - else: - this_alphabet = list(string.ascii_lowercase) - random.shuffle(this_alphabet) - this_alphabet = cat(this_alphabet) - worker_args.append((ciphertext, this_alphabet, max_iterations, fitness)) + if not plain_alphabet: + plain_alphabet = string.ascii_lowercase + if not cipher_alphabet: + cipher_alphabet = list(string.ascii_lowercase) + random.shuffle(cipher_alphabet) + cipher_alphabet = cat(cipher_alphabet) + worker_args.append((ciphertext, plain_alphabet, cipher_alphabet, + initial_temperature, max_iterations, fitness)) with Pool() as pool: - breaks = pool.starmap(monoalphabetic_break_hillclimbing_worker, + breaks = pool.starmap(simulated_annealing_break_worker, worker_args, chunksize) return max(breaks, key=lambda k: k[1]) -def monoalphabetic_break_hillclimbing_worker(message, alphabet, - max_iterations, fitness): + +def simulated_annealing_break_worker(message, plain_alphabet, cipher_alphabet, + t0, max_iterations, fitness): def swap(letters, i, j): if i > j: i, j = j, i @@ -246,17 +333,56 @@ def monoalphabetic_break_hillclimbing_worker(message, alphabet, else: return (letters[:i] + letters[j] + letters[i+1:j] + letters[i] + letters[j+1:]) - best_alphabet = alphabet - best_fitness = float('-inf') + + temperature = t0 + + dt = t0 / (0.9 * max_iterations) + + current_alphabet = cipher_alphabet + alphabet = current_alphabet + cipher_translation = ''.maketrans(current_alphabet, plain_alphabet) + plaintext = message.translate(cipher_translation) + current_fitness = fitness(plaintext) + + best_alphabet = current_alphabet + best_fitness = current_fitness + best_plaintext = plaintext + + # print('starting for', max_iterations) for i in range(max_iterations): - alphabet = swap(alphabet, random.randrange(26), random.randrange(26)) - cipher_translation = ''.maketrans(string.ascii_lowercase, alphabet) + swap_a = random.randrange(26) + swap_b = (swap_a + int(random.gauss(0, 4))) % 26 + alphabet = swap(current_alphabet, swap_a, swap_b) + cipher_translation = ''.maketrans(alphabet, plain_alphabet) plaintext = message.translate(cipher_translation) - if fitness(plaintext) > best_fitness: - best_fitness = fitness(plaintext) - best_alphabet = alphabet - print(i, best_alphabet, best_fitness, plaintext) - return best_alphabet, best_fitness + new_fitness = fitness(plaintext) + try: + sa_chance = math.exp((new_fitness - current_fitness) / temperature) + except (OverflowError, ZeroDivisionError): + # print('exception triggered: new_fit {}, current_fit {}, temp {}'.format(new_fitness, current_fitness, temperature)) + sa_chance = 0 + if (new_fitness > current_fitness or random.random() < sa_chance): + # logger.debug('Simulated annealing: iteration {}, temperature {}, ' + # 'current alphabet {}, current_fitness {}, ' + # 'best_plaintext {}'.format(i, temperature, current_alphabet, + # current_fitness, best_plaintext[:50])) + + # logger.debug('new_fit {}, current_fit {}, temp {}, sa_chance {}'.format(new_fitness, current_fitness, temperature, sa_chance)) + current_fitness = new_fitness + current_alphabet = alphabet + + if current_fitness > best_fitness: + best_alphabet = current_alphabet + best_fitness = current_fitness + best_plaintext = plaintext + if i % 500 == 0: + logger.debug('Simulated annealing: iteration {}, temperature {}, ' + 'current alphabet {}, current_fitness {}, ' + 'best_plaintext {}'.format(i, temperature, current_alphabet, + current_fitness, plaintext[:50])) + temperature = max(temperature - dt, 0.001) + + return best_alphabet, best_fitness # current_alphabet, current_fitness def vigenere_keyword_break_mp(message, wordlist=keywords, fitness=Pletters, @@ -301,7 +427,7 @@ def vigenere_frequency_break(message, max_key_length=20, fitness=Pletters): """ def worker(message, key_length, fitness): splits = every_nth(sanitised_message, key_length) - key = cat([chr(caesar_break(s)[0] + ord('a')) for s in splits]) + key = cat([unpos(caesar_break(s)[0]) for s in splits]) plaintext = vigenere_decipher(message, key) fit = fitness(plaintext) return key, fit @@ -376,8 +502,7 @@ def beaufort_variant_frequency_break(message, max_key_length=20, fitness=Pletter """ def worker(message, key_length, fitness): splits = every_nth(sanitised_message, key_length) - key = cat([chr(-caesar_break(s)[0] % 26 + ord('a')) - for s in splits]) + key = cat([unpos(-caesar_break(s)[0]) for s in splits]) plaintext = beaufort_variant_decipher(message, key) fit = fitness(plaintext) return key, fit @@ -729,6 +854,90 @@ def bifid_break_worker(message, keyword, wrap_alphabet, period, fitness): return (keyword, wrap_alphabet, period), fit +def autokey_sa_break( message + , min_keylength=2 + , max_keylength=20 + , workers=10 + , initial_temperature=200 + , max_iterations=20000 + , fitness=Pletters + , chunksize=1 + , result_count=1 + ): + """Break an autokey cipher by simulated annealing + """ + worker_args = [] + ciphertext = sanitise(message) + for keylength in range(min_keylength, max_keylength+1): + for i in range(workers): + key = cat(random.choice(string.ascii_lowercase) for _ in range(keylength)) + worker_args.append((ciphertext, key, + initial_temperature, max_iterations, fitness)) + + with Pool() as pool: + breaks = pool.starmap(autokey_sa_break_worker, + worker_args, chunksize) + if result_count <= 1: + return max(breaks, key=lambda k: k[1]) + else: + return sorted(set(breaks), key=lambda k: k[1], reverse=True)[:result_count] + + +def autokey_sa_break_worker(message, key, + t0, max_iterations, fitness): + + temperature = t0 + + dt = t0 / (0.9 * max_iterations) + + plaintext = autokey_decipher(message, key) + current_fitness = fitness(plaintext) + current_key = key + + best_key = current_key + best_fitness = current_fitness + best_plaintext = plaintext + + # print('starting for', max_iterations) + for i in range(max_iterations): + swap_pos = random.randrange(len(current_key)) + swap_char = random.choice(string.ascii_lowercase) + + new_key = current_key[:swap_pos] + swap_char + current_key[swap_pos+1:] + + plaintext = autokey_decipher(message, new_key) + new_fitness = fitness(plaintext) + try: + sa_chance = math.exp((new_fitness - current_fitness) / temperature) + except (OverflowError, ZeroDivisionError): + # print('exception triggered: new_fit {}, current_fit {}, temp {}'.format(new_fitness, current_fitness, temperature)) + sa_chance = 0 + if (new_fitness > current_fitness or random.random() < sa_chance): + # logger.debug('Simulated annealing: iteration {}, temperature {}, ' + # 'current alphabet {}, current_fitness {}, ' + # 'best_plaintext {}'.format(i, temperature, current_alphabet, + # current_fitness, best_plaintext[:50])) + + # logger.debug('new_fit {}, current_fit {}, temp {}, sa_chance {}'.format(new_fitness, current_fitness, temperature, sa_chance)) +# print(new_fitness, new_key, plaintext[:100]) + current_fitness = new_fitness + current_key = new_key + + if current_fitness > best_fitness: + best_key = current_key + best_fitness = current_fitness + best_plaintext = plaintext + if i % 500 == 0: + logger.debug('Simulated annealing: iteration {}, temperature {}, ' + 'current key {}, current_fitness {}, ' + 'best_plaintext {}'.format(i, temperature, current_key, + current_fitness, plaintext[:50])) + temperature = max(temperature - dt, 0.001) + +# print(best_key, best_fitness, best_plaintext[:70]) + return best_key, best_fitness # current_alphabet, current_fitness + + def pocket_enigma_break_by_crib(message, wheel_spec, crib, crib_position): """Break a pocket enigma using a crib (some plaintext that's expected to be in a certain position). Returns a list of possible starting wheel