'''
Created on 22 de ago de 2016
@author: paic-elloa-janderson
'''
from perceptron import Perceptron
if __name__ == '__main__':
p = Perceptron()
p.setValues()
n = 0
while(n < p.getQtdInput()):
p.calculateU(n)
p.calculateOutput()
p.calculateError(n)
if(p.getError() != 0):
p.adjustWeigth(n)
print("Pesos: ",p.weigth)
n = -1
n = n + 1
def trainSingle(config, trainingInput, labels, trainingInputTest, labelsTest):
# Create with shape because of N points of M components being NxM
perceptron = Perceptron(trainingInput.shape[1], config.activation, config.learningRate, config.beta, config.momentum, config.alpha)
# For graphing
weights.append(perceptron.getWeights())
# Define output file and prepare output
filenameWeights = OUTPUT_DIR + ('run_input%s_%s_%s_%s.csv' % (trainingInput.shape[0], config.activation, config.learningRate, time.time()))
filenameErrors = OUTPUT_DIR + ('run_errors%s_%s_%s_%s.csv' % (trainingInput.shape[0], config.activation, config.learningRate, time.time()))
filenameTrain = OUTPUT_DIR + ('run_train%s_%s_%s_%s.csv' % (trainingInput.shape[0], config.activation, config.learningRate, time.time()))
filenameTest = OUTPUT_DIR + ('run_test%s_%s_%s_%s.csv' % (trainingInput.shape[0], config.activation, config.learningRate, time.time()))
prepareOutput(filenameWeights, OUTPUT_WEIGHTS_FIELDNAMES)
prepareOutput(filenameErrors, OUTPUT_ERRORS_FIELDNAMES)
prepareOutput(filenameTrain, OUTPUT_METRICS_FIELDNAMES if config.calculateMetrics else OUTPUT_ACCURACY_FIELDNAMES)
prepareOutput(filenameTest, OUTPUT_METRICS_FIELDNAMES if config.calculateMetrics else OUTPUT_ACCURACY_FIELDNAMES)
trainingSize = trainingInput.shape[0]
iterations = 0
error = 1
try:
while iterations < config.iterations and error > config.error:
stdout.write("Epoch #" + str(iterations) + "\r")
stdout.flush()
# Getting a random index order
indexes = getRandomDatasetOrder(trainingSize)
for x_i in indexes:
# Make the prediction
summation, prediction = predict(perceptron, trainingInput[x_i])
# Correct Perceptron weights.
perceptron.correctWeights(trainingInput[x_i], labels[x_i], prediction, summation)
# Calculate error.
error = 0
for inputs, label in zip(trainingInput, labels):
summation, prediction = predict(perceptron, inputs)
error += perceptron.calculateError(label, prediction)
errors.append(error[0])
weights.append(perceptron.getWeights())
# Calculate metrics
accuracy, errorMetric, precision, recall, f1 = testPerceptron(perceptron, trainingInput, labels, config.delta, calculateMetrics=config.calculateMetrics)
trainMetrics.append([accuracy, precision, recall, f1] if config.calculateMetrics else [accuracy])
accuracy, errorMetric, precision, recall, f1 = testPerceptron(perceptron, trainingInputTest, labelsTest, config.delta, calculateMetrics=config.calculateMetrics)
testMetrics.append([accuracy, precision, recall, f1] if config.calculateMetrics else [accuracy])
iterations += 1
# Write output
with open(filenameWeights, 'a') as csv_file:
# Get instance of the writer
csv_writer = csv.DictWriter(csv_file, fieldnames=OUTPUT_WEIGHTS_FIELDNAMES)
writeAllWeights(csv_writer)
# Write errors
with open(filenameErrors, 'a') as csv_file:
# Get instance of the writer
csv_writer = csv.DictWriter(csv_file, fieldnames=OUTPUT_ERRORS_FIELDNAMES)
writeAllErrors(csv_writer)
print("\n\n######################\nTESTING\n######################")
print("Training Results:")
accuracy, error, precision, recall, f1 = testPerceptron(perceptron, trainingInput, labels, config.delta, True, calculateMetrics=config.calculateMetrics)
printMetrics(accuracy, error, precision, recall, f1)
with open(filenameTrain, 'a') as csv_file:
# Get instance of the writer
csv_writer = csv.DictWriter(csv_file, fieldnames=OUTPUT_METRICS_FIELDNAMES if config.calculateMetrics else OUTPUT_ACCURACY_FIELDNAMES)
if config.calculateMetrics:
writeAllMetrics(csv_writer, trainMetrics)
else:
writeAccuracyMetrics(csv_writer, trainMetrics)
print("Learning Results:")
accuracy, error, precision, recall, f1 = testPerceptron(perceptron, trainingInputTest, labelsTest, config.delta, True, calculateMetrics=config.calculateMetrics)
printMetrics(accuracy, error, precision, recall, f1)
with open(filenameTest, 'a') as csv_file:
# Get instance of the writer
csv_writer = csv.DictWriter(csv_file, fieldnames=OUTPUT_METRICS_FIELDNAMES if config.calculateMetrics else OUTPUT_ACCURACY_FIELDNAMES)
if config.calculateMetrics:
writeAllMetrics(csv_writer, testMetrics)
else:
writeAccuracyMetrics(csv_writer, testMetrics)
except KeyboardInterrupt:
print("Finishing up...")
