# We collect the various base boosting (hyper)parameters.
options_with_lad = dict(n_regressors=n_regressors,
boosting_loss=ls_loss,
line_search_options=line_search_options_with_lad)
options_with_huber = dict(n_regressors=n_regressors,
boosting_loss=huber_loss,
line_search_options=line_search_options_with_huber)
print('Building scoring DataFrame for each single-target regression subtask.')
test_error = []
for index in range(5):
if index in [0, 1, 4]:
# We make an instance of Regressor with our choice of stacking
# for the single-target regression subtasks: 1, 2, and 5.
reg = Regressor(regressor_choice=stacking_basis_fn,
params=paper_params(index),
target_index=index,
stacking_options=dict(layers=stack),
base_boosting_options=options_with_huber)
elif index == 3:
# We make an instance of Regressor with our choice of stacking
# for the single-target regression subtask: 4.
reg = Regressor(regressor_choice=stacking_basis_fn,
params=paper_params(index),
target_index=index,
stacking_options=dict(layers=stack),
base_boosting_options=options_with_lad)
else:
# We make an instance of Regressor with our choice of ridge
# regression for the single-target regression subtask: 3.
# The parameter alpha denotes the regularization strength
# We collect the various base boosting (hyper)parameters.
base_boosting_options = dict(n_regressors=n_regressors,
boosting_loss=boosting_loss,
line_search_options=line_search_options)
# We focus on the third single-target regression subtask,
# as this is the least difficult subtask for the base regressor.
# Note that the index corresponds to the Python convention.
index = 2
# We generate the augmented learning curve with our choice
# of stacking in the single-target regression subtask: 3,
# wherein we use 40 different training data sizes. The
# first training data size corresponds to a quarter of
# the training data, and the last training data size
# corresponds to the full amount of training data.
plot_learning_curve(regressor_choice=basis_fn,
title='Augmented learning curve',
X=X_train, y=y_train, verbose=1, cv=5,
train_sizes=np.linspace(0.25, 1.0, 40),
alpha=0.1, train_color='b',
cv_color='orange', y_ticks_step=0.15,
fill_std=False, legend_loc='best',
save_plot=False, path=None,
pipeline_transform='quantilenormal',
pipeline_memory=None, chain_order=None,
params=paper_params(index), target_index=index,
ylabel='Mean absolute error', stacking_options=dict(layers=stack),
base_boosting_options=base_boosting_options,
return_incumbent_score=True)