def predict_function(x, y, model):
self.avg_error = float(model.sum_error) / float(model.i)
if self.output:
print "Average error: ", self.avg_error, "\n"
X_array = np.array(x).reshape(1, len(x))
y_array = np.array(y).reshape(1, len(y))
return evaluate_error(X_array, y_array, model.w) ** 0.5
def predict_function(x, y, model):
self.avg_error = float(model.sum_error) / float(model.i)
if self.output:
print "Average error: ", self.avg_error, "\n"
X_array = np.array(x).reshape(1, len(x))
y_array = np.array(y).reshape(1, len(y))
return evaluate_error(X_array, y_array, model.w)**0.5
def train_function(x, y, model, window_state):
if not model:
class Model:
w = np.zeros((x.shape[1] + 1, 1))
sum_error = 0
i = 0
model = Model()
model.w = train(x, y, self.draw)
self.w = model.w
error = evaluate_error(x, y, model.w)
if self.output:
print "Error: ", error
model.sum_error += error
model.i += 1
return model
def train_function(x, y, model, window_state):
if not model:
class Model:
w = np.zeros((x.shape[1] + 1, 1))
sum_error = 0
i = 0
model = Model()
model.w = train(x, y, self.draw)
self.w = model.w
error = evaluate_error(x, y, model.w)
if self.output:
print "Error: ", error
model.sum_error += error
model.i += 1
return model
def train_function(x, y, model, window_state):
step_size = window_state[3]
current_window_size = window_state[0]
max_window_size = window_state[4]
# Initialize model if not initialized
if not model:
class Model:
w = np.zeros((x.shape[1] + 1, 1))
sum_error = 0
i = 0
# Use incremental matrix state if data has 1 feature
if x.shape[1] == 1:
x_sum = 0
y_sum = 0
xy_sum = 0
xx_sum = 0
model = Model()
# Set incremental matrix state if data has 1 feature
if x.shape[1] == 1:
model.x_sum = np.sum(x)
model.y_sum = np.sum(y)
model.xy_sum = np.sum(x * y)
model.xx_sum = np.sum(x ** 2)
# Model is already initialized
# If data has 1 feature, add last step_size points from
# sums
elif x.shape[1] == 1:
for i in range(-step_size, 0):
x_value = x[i].tolist()[0]
y_value = y[i].tolist()[0]
model.x_sum += x_value
model.y_sum += y_value
model.xy_sum += x_value * y_value
model.xx_sum += x_value ** 2
# If data has 1 feature, compute w with incremental matrix
if x.shape[1] == 1:
n = x.shape[0]
model.w[1] = (model.xy_sum - model.x_sum * model.y_sum /
float(n)) / (model.xx_sum - model.x_sum *
model.x_sum / float(n))
model.w[0] = (model.y_sum/float(n) - model.w[1] *
model.x_sum/float(n))
self.w = model.w
if self.draw:
plot(x, y, model.w)
# If the window has not reached steady state and the next
# window will be at steady state, remove points to
# correctly update the sums
if (max_window_size - current_window_size < step_size and
not window_state[1]):
for i in range(0, step_size - (max_window_size -
current_window_size)):
x_value = x[i].tolist()[0]
y_value = y[i].tolist()[0]
model.x_sum -= x_value
model.y_sum -= y_value
model.xy_sum -= x_value * y_value
model.xx_sum -= x_value ** 2
# If the window has reached steady state, remove the first
# step size points from sums
if window_state[1]:
for i in range(0, step_size):
x_value = x[i].tolist()[0]
y_value = y[i].tolist()[0]
model.x_sum -= x_value
model.y_sum -= y_value
model.xy_sum -= x_value * y_value
model.xx_sum -= x_value ** 2
# The data has more than 1 feature, train using SGD
else:
model.w = train_sgd(x, y, self.alpha, model.w)
if self.draw:
plot(x, y, model.w)
self.w = model.w
error = evaluate_error(x, y, model.w)
if self.output:
print "Error: ", error
model.sum_error += error
model.i += 1
return model
def train_function(x, y, model, window_state):
step_size = window_state[3]
current_window_size = window_state[0]
max_window_size = window_state[4]
# Initialize model if not initialized
if not model:
class Model:
w = np.zeros((x.shape[1] + 1, 1))
sum_error = 0
i = 0
# Use incremental matrix state if data has 1 feature
if x.shape[1] == 1:
x_sum = 0
y_sum = 0
xy_sum = 0
xx_sum = 0
model = Model()
# Set incremental matrix state if data has 1 feature
if x.shape[1] == 1:
model.x_sum = np.sum(x)
model.y_sum = np.sum(y)
model.xy_sum = np.sum(x * y)
model.xx_sum = np.sum(x**2)
# Model is already initialized
# If data has 1 feature, add last step_size points from
# sums
elif x.shape[1] == 1:
for i in range(-step_size, 0):
x_value = x[i].tolist()[0]
y_value = y[i].tolist()[0]
model.x_sum += x_value
model.y_sum += y_value
model.xy_sum += x_value * y_value
model.xx_sum += x_value**2
# If data has 1 feature, compute w with incremental matrix
if x.shape[1] == 1:
n = x.shape[0]
model.w[1] = (
model.xy_sum - model.x_sum * model.y_sum / float(n)
) / (model.xx_sum - model.x_sum * model.x_sum / float(n))
model.w[0] = (model.y_sum / float(n) -
model.w[1] * model.x_sum / float(n))
self.w = model.w
if self.draw:
plot(x, y, model.w)
# If the window has not reached steady state and the next
# window will be at steady state, remove points to
# correctly update the sums
if (max_window_size - current_window_size < step_size
and not window_state[1]):
for i in range(
0, step_size -
(max_window_size - current_window_size)):
x_value = x[i].tolist()[0]
y_value = y[i].tolist()[0]
model.x_sum -= x_value
model.y_sum -= y_value
model.xy_sum -= x_value * y_value
model.xx_sum -= x_value**2
# If the window has reached steady state, remove the first
# step size points from sums
if window_state[1]:
for i in range(0, step_size):
x_value = x[i].tolist()[0]
y_value = y[i].tolist()[0]
model.x_sum -= x_value
model.y_sum -= y_value
model.xy_sum -= x_value * y_value
model.xx_sum -= x_value**2
# The data has more than 1 feature, train using SGD
else:
model.w = train_sgd(x, y, self.alpha, model.w)
if self.draw:
plot(x, y, model.w)
self.w = model.w
error = evaluate_error(x, y, model.w)
if self.output:
print "Error: ", error
model.sum_error += error
model.i += 1
return model
