def __init__(self):
self.feature_vec = [
features.LinearX1(),
features.LinearX2(),
features.SquareX1(),
features.ExpX2(),
features.LogX1(),
features.Identity()
]
self.feature_weights = [1, 2, 1, 0.1, 10, 40]
self.noise_model = noise.NoiseModel()
self.max_x1 = 10
self.max_x2 = 10
self.saver = saver.DataSaver('data', 'data_samples.pkl')
# data_generator = data.DataGenerator()
data_saver = saver.DataSaver('data', 'data_samples.pkl')
input_data, output_data = data_saver.restore_from_file()
n_samples = input_data.shape[0]
# Split data into training and validation
ratio_train_validate = 0.8
idx_switch = int(n_samples * ratio_train_validate)
training_input = input_data[:idx_switch, :]
training_output = output_data[:idx_switch, :]
validation_input = input_data[idx_switch:, :]
validation_output = output_data[idx_switch:, :]
# Fit model
lm = model.LinearRegressionModel()
lm.set_feature_vector([features.LinearX1(), #TODO ])
lm.fit(training_input, training_output)
# sklearn
#TODO
# Validation
mse = lm.validate(validation_input, validation_output)
print('MSE: {}'.format(mse))
print(' ')
print('feature weights \n{}'.format(lm.beta))
validation_predictions = lm.predict(validation_input)
feature_matrix_validation = lm.compute_feature_matrix(validation_input)
validation_predictions_sk = linear_regression_sk.predict(
feature_matrix_validation)
mse = 0
# Fit model for each part
for part in range(0, n_parts):
train_start_i = int(part * samp_per_part)
test_start_i = int(train_start_i + samp_per_part * ratio_train_validate)
test_end_i = int(train_start_i + samp_per_part)
training_input[part] = input_data[train_start_i:test_start_i, :]
training_output[part] = output_data[train_start_i:test_start_i, :]
validation_input[part] = input_data[test_start_i:test_end_i, :]
validation_output[part] = output_data[test_start_i:test_end_i, :]
lm[part] = model.LinearRegressionModel()
# TODO use and select the new features
lm[part].set_feature_vector([
features.LinearX1(),
features.LinearX2(),
features.LinearX3(),
features.LinearX4(),
features.SquareX1(),
features.SquareX2(),
features.SquareX3(),
features.SquareX4(),
features.ExpX1(),
features.ExpX2(),
features.ExpX3(),
features.ExpX4(),
features.LogX1(),
features.LogX2(),
features.LogX3(),
features.LogX4(),
