def trainNeuron(df, maxError = 0.4, maxIter = 1000):
neuron = Perceptron()
neuron.af = af.linear
acc = 0
iter = 0
range = 5
while (1 - acc > maxError) and (iter < maxIter):
# Generate random weights and biases:
neuron.weights = (rd.uniform(-range, range), rd.uniform(-range,range))
neuron.bias = rd.uniform(-range, range)
# Calculate accuracy
acc = calculateAccuracy(df, neuron)
iter += 1
return {'weights': tuple(neuron.weights), 'bias': neuron.bias, 'acc': acc, 'iter': iter}
def trainPerceptron(df, minLearningRate=0.005, maxIter=100):
neuron = Perceptron()
acc = 0
iter = 0
learningRate = 1
lastAccuracies = list()
for i in range(10):
lastAccuracies.append(0)
lastAccuracies.append(1)
# Track error during training
errorEvolution = list()
# Generate random weights and biases:
radius = 5
neuron.weights = list()
for i in range(df.shape[1] - 1):
neuron.weights.append(rd.uniform(-radius, radius))
#rd.uniform(-radius, radius), rd.uniform(-radius,radius)]
neuron.bias = rd.uniform(-radius, radius)
while (learningRate > minLearningRate) and (iter < maxIter):
# Loop over dataset once, update weights and return accuracy
acc = iteratePerceptronTraining(df, neuron)
# Update learningRate
del lastAccuracies[0]
lastAccuracies.append(acc)
learningRate = max(lastAccuracies) - min(lastAccuracies)
# Update error
errorEvolution.append(1 - acc)
# Update iteration counter
iter += 1
return neuron, errorEvolution