# Initial weighting---uniform
original_weights_train = numpy.ones(len(original_train))
original_weights_test = numpy.ones(len(original_test))
# Plot settings
hist_settings = {'bins': 50, 'density': True, 'alpha': 0.7}
################
# Plot initial #
################
draw_distributions('initial.png',
columns, original, target, original_weights,
filename_as_title=True,
xlim=((0, 50), (0, 12)),
ylim=((0, 0.07), (0, 1.7)),
nrows=1, ncols=2, hist_settings=hist_settings)
draw_distributions('initial_train.png',
columns, original_train, target_train, original_weights_train,
filename_as_title=True,
xlim=((0, 50), (0, 12)),
ylim=((0, 0.07), (0, 1.7)),
nrows=1, ncols=2, hist_settings=hist_settings)
draw_distributions('initial_test.png',
columns, original_test, target_test, original_weights_test,
filename_as_title=True,
xlim=((0, 50), (0, 12)),
ylim=((0, 0.07), (0, 1.7)),
###############################
reweighter = reweight.GBReweighter(n_estimators=500,
learning_rate=0.1,
max_depth=3,
min_samples_leaf=1000,
gb_args={'subsample': 0.4})
reweighter.fit(original_train, target_train)
gb_weights_test = reweighter.predict_weights(original_test)
# Validate reweighting rule on the test part comparing 1d projections
draw_distributions('gbr5_validate_ConeMult.png', [
columns[0],
],
original_test,
target_test,
gb_weights_test,
filename_as_title=True,
nrows=1,
ncols=1,
hist_settings=hist_settings)
draw_distributions('gbr5_validate_VtxMult.png', [columns[1]],
original_test,
target_test,
gb_weights_test,
filename_as_title=True,
nrows=1,
ncols=1,
hist_settings=hist_settings)
draw_distributions('gbr5_validate_ConePtAsym.png', [columns[2]],
# Divide target samples into training ant test parts
target_train, target_test = train_test_split(target)
original_weights_train = numpy.ones(len(original_train))
original_weights_test = numpy.ones(len(original_test))
# Pay attention, actually we have very few data
print('Length of original data: %s\nLength of target data: %s' %
(len(original), len(target)))
###################################################
# Print the unmodified original and test data set #
###################################################
draw_distributions('initial.png', columns, original, target, original_weights)
print_statistics(columns, original, target, original_weights)
# Train part of original distribution
draw_distributions('initial_train.png', columns, original_train, target_train,
original_weights_train)
# Test part of target distribution
draw_distributions('initial_test.png', columns, original_test, target_test,
original_weights_test)
###############################
# Gradient boosted Reweighter #
###############################
reweighter = reweight.GBReweighter(n_estimators=50,
reweighter = reweight.GBReweighter(n_estimators=50,
learning_rate=0.1,
max_depth=3,
min_samples_leaf=1000,
gb_args={'subsample': 0.4})
reweighter.fit(original_train, target_train)
gb_weights_test = reweighter.predict_weights(original_test)
# Validate reweighting rule on the test part comparing 1d projections
draw_distributions('gb_weights_test.png',
columns,
original_test,
target_test,
gb_weights_test,
filename_as_title=True,
xlim=((0, 50), (0, 12)),
ylim=((0, 0.10), (0, 3.0)),
nrows=1,
ncols=2,
hist_settings=hist_settings)
######################
# Folding Reweighter #
######################
# Define base reweighter
reweighter_base = reweight.GBReweighter(n_estimators=50,
learning_rate=0.1,
max_depth=3,
min_samples_leaf=1000,
)
reweighter = reweight.FoldingReweighter(reweighter_base, n_folds=2)
# Not need to divide data into train/test parts
reweighter.fit(original, target, target_weight=target_weights)
# Prediect weights for the input file
folding_weights = reweighter.predict_weights(toReweight)
draw_distributions(
'GBR4_validate.png',
columns,
toReweight,
target,
folding_weights,
filename_as_title=True,
# yscale=('log',),
nrows=2,
ncols=2,
hist_settings=hist_settings)
draw_distributions('GBR4_before_after.png',
columns,
toReweight,
toReweight,
None,
folding_weights,
filename_as_title=True,
yscale=('linear', 'linear', 'linear', 'linear'),
nrows=2,