def set_uniform_config(self, ca_rule=105, R=4, C=3, I=12, classifier="linear-svm",
encoding="random_mapping", time_transition="random_permutation"):
# sets up elementary CA:
self.reservoir = ca.ElemCAReservoir()
self.reservoir.set_uniform_rule(ca_rule)
# clf
if classifier=="linear-svm":
self.classifier = scikit_clfs.SVM()
elif classifier =="perceptron_sgd":
self.classifier = scikit_clfs.SGD()
elif classifier =="linear_regression":
self.classifier = scikit_clfs.LinReg()
elif classifier =="tlf_ann":
self.classifier = tflann.ANN()
# Encoder
if encoding == "random_mapping":
self.encoder = enc.RandomMappingEncoder()
self.encoder.R = R
self.encoder.C = C
self.I = I
if time_transition=="normalized_addition":
self.time_transition = time_trans.RandomAdditionTimeTransition()
elif time_transition == "random_permutation":
self.time_transition = time_trans.RandomPermutationTransition()
elif time_transition == "xor":
self.time_transition = time_trans.XORTimeTransition()
def set_parallel_reservoir_config(self,
ca_rules=(105, 110),
parallel_size_policy="unbounded",
R=4,
C=3,
I=12,
classifier="linear-svm",
encoding="random_mapping",
time_transition="normalized_addition"):
# sets up elementary CA:
self.reservoir = ca.ElemCAReservoir()
self.reservoir.set_rules(ca_rules)
#if parallel_size_policy
self.parallelizer = prl.ParallelNonUniformEncoder(
self.reservoir.rules, parallel_size_policy)
# clf
if classifier == "linear-svm":
self.classifier = svm.SVM()
elif classifier == "tlf_ann":
self.classifier = tflann.ANN()
# Encoder
if encoding == "random_mapping":
self.encoder = rnd_map.RandomMappingEncoder(self.parallelizer)
self.encoder.R = R
self.encoder.C = C
self.I = I
if time_transition == "normalized_addition":
self.time_transition = norm_add.RandomAdditionTimeTransition()
elif time_transition == "random_permutation":
self.time_transition = rnd_perm.RandomPermutationTransition()
def calculate_pseudo_lambda(self, rule_set):
ca_simulator = ca.ElemCAReservoir()
ca_simulator.set_rules(rule_set)
initial_input = np.zeros(len(rule_set))
initial_input[int(len(initial_input)/2)] = 1
simulation = ca_simulator.run_simulation(initial_input, 1)
second_row = simulation[1]
ones = list(second_row).count(1)
return int(1000/(ones+1))
def set_uniform_margem_config(self, rule_scheme=None,N=4, R_i=4, R=20, I=4, classifier="perceptron_sgd", time_transition="xor"):
"""
:param rule:
:param R_i:
:param R: Padding on each side of the input (buffers)
:param I:
:param classifier:
:param time_transition:
:return:
"""
print("Running rotation with values: rule: " + str("unkn. ") + ", R_i: " + str(R_i) + ", R:" + str(R), "I: " +str(I))
ca_size = N * R_i + R*2 # Used to create rule scheme
# sets up elementary CA:
before_time = time.time()
self.reservoir = ca.ElemCAReservoir(ca_size)
self.reservoir.set_rule_config(rule_scheme)
#self.reservoir.set_rules(rule)
#self.reservoir.set_uniform_rule(rule)
# clf
if classifier=="linear-svm":
self.classifier = scikit_clfs.SVM()
elif classifier =="perceptron_sgd":
self.classifier = scikit_clfs.SGD()
elif classifier == "random_forest":
self.classifier = scikit_clfs.RandomForest()
elif classifier =="linear_regression":
self.classifier = scikit_clfs.LinReg()
elif classifier =="tlf_ann":
self.classifier = tflann.ANN()
# Encoder
self.encoder = enc.RotationEncoder()
self.encoder.R = R
self.encoder.R_i = R_i
self.I = I
if time_transition == "normalized_addition":
self.time_transition = time_trans.RandomAdditionTimeTransition()
elif time_transition == "random_permutation":
self.time_transition = time_trans.RandomPermutationTransition()
elif time_transition == "xor":
self.time_transition = time_trans.XORTimeTransition()
elif time_transition == "or":
self.time_transition = time_trans.ORTimeTransition()
elif time_transition == "and":
self.time_transition = time_trans.ANDTimeTransition()
elif time_transition == "nand":
self.time_transition = time_trans.NANDTimeTransition()
def rule_vizualize(self, ca_rule, R, I, _input):
width = 100
gens = I
elem_ca = ca.ElemCAReservoir()
elem_ca.set_rule(ca_rule)
short_input = _input[:]
for _ in range(R):
_input.extend(short_input)
output = elem_ca.run_simulation([_input], gens)
self.visualize(output)
def set_non_uniform_config(self, rule_scheme, R=6, C=4, I=4, classifier="linear-svm", encoding="random_mapping",
time_transition="random_permutation"):
"""
:param rule_scheme: Must be a way to design a non-uniform ca reservoir of the exact same size (R*C*N etc)
:param R:
:param C:
:param I:
:param classifier:
:param encoding:
:param time_transition:
:return:
"""
# sets up elementary CA:
self.reservoir = ca.ElemCAReservoir()
self.reservoir.set_rules(rule_scheme.get_scheme(4*R*C))
#self.reservoir.set_uniform_rule(90)
# clf
if classifier=="linear-svm":
self.classifier = scikit_clfs.SVM()
elif classifier =="perceptron_sgd":
self.classifier = scikit_clfs.SGD()
elif classifier =="linear_regression":
self.classifier = scikit_clfs.LinReg()
elif classifier =="tlf_ann":
self.classifier = tflann.ANN()
# Encoder
if encoding == "random_mapping":
self.encoder = enc.RandomMappingEncoder()
self.encoder.R = R
self.encoder.C = C
self.I = I
if time_transition=="normalized_addition":
self.time_transition = time_trans.RandomAdditionTimeTransition()
elif time_transition == "random_permutation":
self.time_transition = time_trans.RandomPermutationTransition()
elif time_transition == "xor":
self.time_transition = time_trans.XORTimeTransition()
def set_random_mapping_config(self, ca_rule_scheme=None, N=4, R=4, C=3, I=12, classifier="linear-svm", time_transition="random_permutation", mapping_permutations=True):
#print("Setting config: R:" + str(R) + " C:" +str(C) + " I:" + str(I) + " clf: " + str(classifier) + " rule scheme: " + str(ca_rule_scheme.rule_list))
ca_size = N*R*C # Used to create rule scheme
# sets up elementary CA:
#print(ca_rule_scheme)
self.reservoir = ca.ElemCAReservoir(ca_size)
self.reservoir.set_rule_config(ca_rule_scheme)
# clf
if classifier == "linear-svm":
self.classifier = scikit_clfs.SVM()
elif classifier == "perceptron_sgd":
self.classifier = scikit_clfs.SGD()
elif classifier == "linear_regression":
self.classifier = scikit_clfs.LinReg()
elif classifier == "tlf_ann":
self.classifier = tflann.ANN()
# Encoder
self.encoder = enc.RandomMappingEncoder(permutations=mapping_permutations)
self.encoder.R = R
self.encoder.C = C
self.I = I
if time_transition == "normalized_addition":
self.time_transition = time_trans.RandomAdditionTimeTransition()
elif time_transition == "random_permutation":
self.time_transition = time_trans.RandomPermutationTransition()
elif time_transition == "xor":
self.time_transition = time_trans.XORTimeTransition()
elif time_transition == "or":
self.time_transition = time_trans.ORTimeTransition()
elif time_transition == "and":
self.time_transition = time_trans.ANDTimeTransition()
elif time_transition == "nand":
self.time_transition = time_trans.NANDTimeTransition()
def use_elem_ca(self, rule_number):
"""
"""
self.reservoir = ca.ElemCAReservoir()
self.reservoir.set_rule(rule_number)
self.rc_framework.reservoir = self.reservoir
