def get_data(n, ic=None, apply_subclass=False):
if apply_subclass:
n = cca.get_subclass(n)
if ic == None:
ic = [rd.randrange(0, k) for _ in range(length)]
te = ca.cellularautomata(n, k, r, ic, t, transient)
x_result = cca.power_spectrum(cca.get_fft(te))
y_result = [0 for _ in range(4)]
y_result[cca.wolfram_class(n) - 1] = 1
# Get data
return x_result, y_result
def feedforward_classify_space_rule(k, r, size, total_ics, model_path, output_file, output_encoding='utf-8'):
t = 3*size
transient = 2*size
m = 2*r + 1
rule_size = int(pow(k, pow(k, m)))
train_data_size = total_ics
# Create the Estimator
print("Create the Estimator")
ca_classifier = tf.estimator.Estimator(
model_fn=model_function_with_dropout, model_dir=model_path)
# Predict CA in ray 1 space
with open(output_file, mode='w', encoding=output_encoding) as a_file:
for n in range(rule_size):
print("Rule:", n)
eval_data = make_set_pair_ne_label(train_data_size, k, r, t, transient)
eval_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": np.array(eval_data[0], dtype=np.float32)},
num_epochs=1,
shuffle=False)
result_predict = ca_classifier.predict(eval_input_fn, predict_keys=None, hooks=None)
eval_prob = [p['probabilities'] for p in result_predict]
eval_prob = [np.sum(line) for line in np.transpose(eval_prob)]
eval_prob = eval_prob / np.sum(eval_prob)
a_file.write(str(n))
a_file.write("\t")
a_file.write(str(cca.wolfram_class(n)))
a_file.write("\t")
a_file.write(str(np.argmax(eval_prob) + 1))
a_file.write("\t")
for val in eval_prob:
a_file.write(str(val))
a_file.write("\t")
a_file.write("\n")
a_file.flush()
def make_set_pair_ne_label(total_ics, k, r, t, transient = 0):
'''
'''
ics = cca.init_conditions_numbers(k, (t - transient), total_ics)
size = (t - transient)
data = []
label = []
rules = []
for ic in ics:
classes = cca.WolframClasses[rnd.randint(0, 3)]
n = classes[rnd.randint(0, len(classes) - 1)]
te = ca.cellularautomata(n, k, r, ca.from_number_fix(ic, k, size), t, transient)
plain_ca = ca.plain(nspec.te2ne(te, k, r))
data.append(plain_ca)
label.append(cca.wolfram_class(n) - 1)
rules.append(n)
return data, label, rules
print("Saving Neural Network...")
save_path = saver.save(sess, checkpoint_prefix)
print("Predicting class...")
print("States: ", k, " Ray: ", r)
with open(checkpoint_prefix + ".txt", "w+") as f:
n = 0
k = 2
r = 1
rule_size = int(np.power(k, np.power(k, 2 * r + 1)))
for n in range(rule_size):
prob = predict_class(sess, n, k, r, prediction)
print(n, "/", rule_size, cca.wolfram_class(n), prob)
f.write(str(n))
f.write(str('\t'))
f.write(str(cca.wolfram_class(n)))
for p in prob:
f.write(str('\t'))
f.write(str(p))
f.write(str('\n'))
if n % 10 == 0:
f.flush()
print()
print("Finish.")