def train(self, data, num_iters):
shouldContinue = True
for iteration_count in xrange(num_iters):
num_errors = 0
for vector, label in data:
activation = dot(self.weights, vector) + self.bias
if label * activation <= 0:
# Update weights
for i in xrange(self.dimensions):
self.weights[i] += label * vector[i]
self.cached_weights[i] += label * vector[i] * self.counter
self.bias += label
self.cached_bias += label * self.counter
num_errors += 1
self.counter += 1
if self.onIteration is not None:
shouldContinue = self.onIteration(iteration_count, self.weights, self.bias, num_errors, len(data))
if shouldContinue is False:
break
return (
array_subtract(self.weights, scale_array(self.cached_weights, 1 / self.counter)),
self.bias - self.cached_bias / self.counter,
)
def get_weights(self):
return array_subtract(self.weights, scale_array(self.cached_weights, 1 / self.counter))
def classify(self, input_vector):
weights = array_subtract(self.weights, scale_array(self.cached_weights, 1 / self.counter))
bias = self.bias - self.cached_bias / self.counter
return sign(dot(weights, input_vector) + bias)
