def train_dnf_network(data, targets, q):
res = MultiOutLNN.train_lnn(
data, targets, 5000 * len(data), len(data[0]), [2**n], 1,
[MultiOutLNN.noisy_and_activation, MultiOutLNN.noisy_or_activation],
False)
wrong = MultiOutLNN.run_lnn(
data, targets, res,
[MultiOutLNN.noisy_and_activation, MultiOutLNN.noisy_or_activation],
False)
er = float(wrong) / float(len(data))
q.put(er)
X_test = np.array(test[0])
Y_test = np.array(test[1])
num_inputs = len(X_train[0])
num_outputs = len(Y_train[0])
## CONFIGURATION ##
hidden_layers = [4]
activations = [
MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_and_activation
]
iterations = len(X_train) * 30
## ## ## ## ## ## ##
res = MultiOutLNN.train_lnn(X_train, Y_train, iterations, num_inputs,
hidden_layers, num_outputs, ["OR", "AND"],
activations, True)
#rule = MultiOutLNN.ExtractRules(len(X_train[0]), res, ["OR", "AND"])
#print(len(rule))
#for i in range(len(rule)):
# print(i)
# print(rule[i])
# print()
print("Training")
print("Total Number Of Samples: ", len(X_train))
print(
"Network Wrong: ",
MultiOutLNN.run_lnn(X_train, Y_train, res, num_inputs, [8], num_outputs,
import numpy as np
dir_path = os.path.dirname(os.path.realpath(__file__))
print(dir_path + "/")
print(tf.gfile.Exists(dir_path + "/"))
mnist = input_data.read_data_sets(dir_path + "/", one_hot=True)
X_train = mnist.train.images
Y_train = mnist.train.labels
X_test = mnist.test.images
Y_test = mnist.test.labels
network = np.load('network.npy')
rules = MultiOutLNN.ExtractFuzzyRules(len(X_train[0]), network, ["OR", "AND"],
0.5, 2)
print()
print()
for i in range(len(rules)):
print(i)
print(rules[i])
print()
##wrong = MultiOutLNN.test_fuzzy_rules(rules, X_train, Y_train)
##er = wrong/len(X_train)
##train_rule_wrong = er
##print("Rules Error Rate: ", er)
wrong = MultiOutLNN.test_fuzzy_rules(rules, X_test, Y_test)
er = wrong / len(X_test)
if i in idx:
X_train.append(data[i])
Y_train.append(targets[i])
else:
X_test.append(data[i])
Y_test.append(targets[i])
return [np.array(X_train), np.array(Y_train)], [np.array(X_test), np.array(Y_test)]
examples, targets = VehicleData.read_data()
train, test = split_data(examples, targets, 0.7)
activations = [MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_and_activation]
net = MultiOutLNN.train_lnn(train[0], train[1], 500 * len(train[0]), len(train[0][0]), [60, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60, 30, 30, 30, 30, 30, 30, 30, 30], 4, activations, True)
print(len(train[0][0]))
print(train[0][0])
rules = MultiOutLNN.ExtractFuzzyRules(len(train[0][0]), net, ['OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'OR', 'AND'], 0.1, 1)
print()
print()
for i in range(len(rules)):
print(i)
#print(len(rules[i]))
print()
print("Training Set")
wrong = MultiOutLNN.run_lnn(train[0], train[1], net, activations, True)
cnf_errors = []
dnf_errors = []
pcep_errors = []
cnf_rule_errors = []
dnf_rule_errors = []
# Peform LOE Cross-Validation
for i in range(0, len(data)):
print(i, " : ", len(data))
data_p = data[np.arange(len(data)) != i]
targets_p = targets[np.arange(len(data)) != i]
print("Training Error")
cnf = MultiOutLNN.train_lnn(
data_p, targets_p, 100000, len(data[0]), [2**len(data[0])],
len(targets[0]),
[MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_and_activation],
True) #train_cnf_network(6, data_p, targets_p, 100000 * 0, 3)
#print("CNF: ", cnf[2])
dnf = MultiOutLNN.train_lnn(
data_p, targets_p, 100000, len(data[0]), [2**len(data[0])],
len(targets[0]),
[MultiOutLNN.noisy_and_activation, MultiOutLNN.noisy_or_activation],
True) #train_dnf_network(6, data_p, targets_p, 100000 * 0, 3)
#print("DNF: ", dnf[2])
pcep = MultiOutNN.train_lnn(data_p, targets_p, 100000, len(data[0]), [30],
len(targets[0]), True)
#print("PCEP: ", pcep[2])
cnf_rule = MultiOutLNN.ExtractRules(len(data[0]), cnf, ["OR", "AND"])
dnf_rule = MultiOutLNN.ExtractRules(len(data[0]), dnf, ["AND", "OR"])
Y_train.append(targets[i])
else:
X_test.append(data[i])
Y_test.append(targets[i])
return X_train, Y_train, X_test, Y_test
acts = [MultiOutLNN.noisy_or_activation, MultiOutLNN.noisy_and_activation]
acts_name = ["AND", "OR"]
targets, examples = read_data()
X_train, Y_train, X_test, Y_test = split_data(examples, targets, 0.7)
res = MultiOutLNN.train_lnn(np.array(X_train), np.array(Y_train), 600000,
len(examples[0]), [60], 3, acts_name, acts, False)
##rule = LNNWithNot.ExtractRules(len(examples[0]), res, acts_name)
##print(rule)
##print("-- Evaluating --")
##print("Training Set, Size = ", len(X_train))
print(
"Network Wrong: ",
MultiOutLNN.run_lnn(X_train, Y_train, res, len(examples[0]), [30], 1, acts,
False))
##print("Rule Wrong: ", LNNWithNot.test(rule[0], X_train, Y_train))
##print()
##print("Testing Set, Size = ", len(X_test))
##print("Network Wrong: ", LNNWithNot.run_lnn(X_test, Y_test, res, len(examples[0]), [30], 1, acts))
##print("Rule Wrong: ", LNNWithNot.test(rule[0], X_test, Y_test))
X_train, Y_train, X_test, Y_test = split_data(data, targets, 0.7)
X_train = X_train[0:55000] / 255
X_test = X_test[0:55000] / 255
num_inputs = len(X_train[0])
num_outputs = len(Y_train[0])
## CONFIGURATION ##
hidden_layers = []
activations = [MultiOutLNN.noisy_and_activation]
iterations = len(X_train) * 30
## ## ## ## ## ## ##
res = MultiOutLNN.train_lnn(X_train, Y_train[0:55000], iterations, num_inputs,
hidden_layers, num_outputs, activations, True)
#rule = MultiOutLNN.ExtractRules(len(X_train[0]), res, ["OR", "AND"])
#print(len(rule))
#for i in range(len(rule)):
# print(i)
# print(rule[i])
# print()
print("Training")
print("Total Number Of Samples: ", len(X_train))
print(
"Network Wrong: ",
MultiOutLNN.run_lnn(X_train, Y_train[0:55000], res, activations, True) /
55000)
import SPECTData
import sys
sys.path.append('../../lib/')
import numpy as np
import MultiOutLNN
train, test = SPECTData.read_data()
activations =[MultiOutLNN.noisy_and_activation, MultiOutLNN.noisy_or_activation]
net = MultiOutLNN.train_lnn(train[0], train[1], 1000 * len(train[0]), len(train[0][0]), [30], 1, [MultiOutLNN.noisy_and_activation, MultiOutLNN.noisy_or_activation], True)
wrong = MultiOutLNN.run_lnn(test[0], test[1], net, activations, True)
er = wrong/len(test[0])
print(er)
if not os.path.exists("./" + name):
os.makedirs(name)
f = open("./{}/{}".format(name, out), 'w+')
f.close()
print()
print("Starting Experement")
#send_notification_via_pushbullet("Experement ({} : {}): {}".format(job_id, task_id, name), "Starting")
for i in range(num):
gc.collect()
print("Starting Iteration: ", i)
with tf.Graph().as_default():
res = MultiOutLNN.train_lnn(X_train, Y_train, iterations, num_inputs,
layers, num_outputs, activations, add_not,
lsm)
training_wrong = MultiOutLNN.run_lnn(X_train, Y_train, res, num_inputs,
layers, num_outputs, activations,
add_not)
testing_wrong = MultiOutLNN.run_lnn(X_test, Y_test, res, num_inputs,
layers, num_outputs, activations,
add_not)
tf.reset_default_graph()
training_error_rate = training_wrong / len(X_train)
testing_error_rate = testing_wrong / len(X_test)
print("Training Error Rate: {}%".format(training_error_rate))
print("Testing Error Rate: {}%".format(testing_error_rate))
import os
import numpy as np
import ReadLensesData
import ConvertData
def to_one_hot(val, m):
res = np.zeros(m)
res[val-1] = 1
return res
data, targets = ReadLensesData.read_data()
data = np.array(data)
targets = np.array([to_one_hot(v, 3) for v in targets])
res = MultiOutLNN.train_lnn(data, targets, int(140000 * 1), len(data[0]), [30], len(targets[0]), [MultiOutLNN.noisy_and_activation, MultiOutLNN.noisy_or_activation], True)
rule = MultiOutLNN.ExtractRules(len(data[0]), res, ["AND", "OR"])
print(len(rule))
for i in range(len(rule)):
print(i)
print(rule[i])
print()
print(data)
print(targets)
print(MultiOutLNN.test(rule, data, targets))
import tensorflow as tf
import random
import numpy as np
import matplotlib.pyplot as plt
import sys
sys.path.append('../../lib/')
import MultiOutLNN
network = np.load('network.npy')
rules = MultiOutLNN.ExtractFuzzyRules(8, network, ['AND', 'OR'], 0.1, 2, False)
print()
print()
for i in range(len(rules)):
print(i)
print(rules[i])
print()
for i in range(len(data)):
#print()
#print(data[i])
ex = np.concatenate([data[i], 1 - data[i]])
#print(ex)
data_prime.append(ex)
data_prime = np.array(data_prime)
res = []
for i in range(1):
print("Experement: ", i, end=' ')
X_train, Y_train, X_test, Y_test = split_data(data_prime, targets, 0.7)
net = MultiOutLNN.train_lnn(X_train, Y_train, iterations, 8,
np.copy(hidden_layers).tolist(), 3,
activations, False)
wrong = MultiOutLNN.run_lnn(X_test, Y_test, net, activations, False)
er = wrong / len(X_test)
print(" -> ", er)
res.append(er)
rules = MultiOutLNN.ExtractFuzzyRules(len(X_train[0]), net, ['OR', 'AND'],
0.5, 2, False)
print()
print()
for i in range(len(rules)):
print(i)
print(rules[i])