def test_LinearRegressionModel_convergence():
# Run until convergence
# assert that model can converge
model = LinearRegressionModel(2, 2)
# NOTE: We use and instead of xor, because xor is non-linear
dataset = datasets.get_and()
model.train(*dataset)
# NOTE: This linear model cannot achieve 0 MSE
assert validation.get_error(model, *dataset) <= 0.1
def test_LinearRegressionModel():
# Run for a couple of iterations
# assert that new error is less than original
model = LinearRegressionModel(2, 2)
# NOTE: We use and instead of xor, because xor is non-linear
dataset = datasets.get_and()
error = validation.get_error(model, *dataset)
model.train(*dataset, iterations=10)
assert validation.get_error(model, *dataset) < error
def test_LinearRegressionModel_get_obj_equals_get_obj_jac_l2_penalty():
_check_get_obj_equals_get_obj_jac(lambda a, o: LinearRegressionModel(
a,
o,
penalty_func=error.L2Penalty(penalty_weight=random.uniform(0.0, 2.0))))
def test_LinearRegressionModel_get_obj_equals_get_obj_jac():
_check_get_obj_equals_get_obj_jac(lambda a, o: LinearRegressionModel(a, o))
def test_LinearRegressionModel_jacobian_l2_penalty():
_check_jacobian(lambda a, o: LinearRegressionModel(
a,
o,
penalty_func=error.L2Penalty(penalty_weight=random.uniform(0.0, 2.0))))
def test_LinearRegressionModel_jacobian():
_check_jacobian(lambda a, o: LinearRegressionModel(a, o))
import functools
from learning import datasets, validation, LinearRegressionModel
from ill import ILL, get_neighborhood_k_nearest
# Grab the popular iris dataset, from 'learning'
dataset = datasets.get_iris()
# Make an underlying model for ILL
# See 'learning' library for more details
underlying_model = LinearRegressionModel(4, 3)
# Make an ILL ensemble of our underlying model
# See code for more options
model = ILL(
underlying_model,
grid_spacing=0.5,
neighborhood_func=functools.partial(
get_neighborhood_k_nearest, k_nearest=5))
# Lets train our ILL
# First, we'll split our dataset into training and testing sets
# Our training set will contain 30 samples from each class
training_set, testing_set = validation.make_train_test_sets(
*dataset, train_per_class=30)
# We could customize training and stopping criteria through
# the arguments of train, but the defaults should be sufficient here
model.train(*training_set)
# Our ILL should converge in a few moments