def objective(params):
print(params)
total_loss = 0
for k in range(4):
URM_train, URM_test, validation_data, test_data = Helper().get_kfold_data(4)[k]
booster = XGBooster(URM_train, validation_data, HybridElasticNetICFUCF)
booster.URM_test = URM_test
booster.fit(train_parameters=deepcopy(params))
loss, _ = Evaluator(test_mode=True).evaluate_recommender_kfold(booster, test_data, sequential=True)
total_loss += loss
total_loss /= 4
print("Map@10 k-fold score:", total_loss)
return -total_loss
from options import Options
import os
import resource
import sys
from xgbooster import XGBooster
if __name__ == '__main__':
# parsing command-line options
options = Options(sys.argv)
# making output unbuffered
if sys.version_info.major == 2:
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)
xgb = XGBooster(options, from_model='../temp/compas_data/compas_data_nbestim_50_maxdepth_3_testsplit_0.2.mod.pkl')
# encode it and save the encoding to another file
xgb.encode()
xgb2 = copy.deepcopy(xgb)
with open('../bench/fairml/compas/compas.samples', 'r') as fp:
lines = fp.readlines()
# timers
ltimes = []
vtimes = []
ftimes = []
etimes = []
if (options.preprocess_categorical):
preprocess_dataset(options.files[0],
options.preprocess_categorical_files)
exit()
if options.files:
xgb = None
if options.train:
data = Data(filename=options.files[0],
mapfile=options.mapfile,
separator=options.separator,
use_categorical=options.use_categorical)
xgb = XGBooster(options, from_data=data)
train_accuracy, test_accuracy, model = xgb.train()
# read a sample from options.explain
if options.explain:
options.explain = [
float(v.strip()) for v in options.explain.split(',')
]
if options.encode:
if not xgb:
xgb = XGBooster(options, from_model=options.files[0])
# encode it and save the encoding to another file
xgb.encode(test_on=options.explain)
import os
import resource
import sys
from xgbooster import XGBooster
if __name__ == '__main__':
# parsing command-line options
options = Options(sys.argv)
# making output unbuffered
if sys.version_info.major == 2:
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)
xgb = XGBooster(
options,
from_model=
'temp/adult_data/adult_data_nbestim_50_maxdepth_3_testsplit_0.2.mod.pkl'
)
# encode it and save the encoding to another file
xgb.encode()
with open('../bench/anchor/adult/adult.samples', 'r') as fp:
lines = fp.readlines()
# timers
atimes = []
vtimes = []
ftimes = []
etimes = []
import os
import resource
import sys
from xgbooster import XGBooster
if __name__ == '__main__':
# parsing command-line options
options = Options(sys.argv)
# making output unbuffered
if sys.version_info.major == 2:
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0)
xgb = XGBooster(
options,
from_model=
'temp/recidivism_data/recidivism_data_nbestim_50_maxdepth_3_testsplit_0.2.mod.pkl'
)
# encode it and save the encoding to another file
xgb.encode()
xgb2 = copy.deepcopy(xgb)
with open('../bench/anchor/recidivism/recidivism.samples', 'r') as fp:
lines = fp.readlines()
# timers
stimes = []
vtimes = []
ftimes = []
def enumerate_all(options, xtype, xnum, smallest, usecld, usemhs, useumcs,
prefix):
# setting the right preferences
options.xtype = xtype
options.xnum = xnum
options.reduce = 'lin'
options.smallest = smallest
options.usecld = usecld
options.usemhs = usemhs
options.useumcs = useumcs
# reading all unique samples
with open('../bench/nlp/spam/quant/spam10.samples', 'r') as fp:
lines = fp.readlines()
# timers and other variables
times, calls = [], []
xsize, exlen = [], []
# doing everything incrementally is expensive;
# let's restart the solver for every 10% of instances
tested = set()
for i, s in enumerate(lines):
if i % (len(lines) / 10) == 0:
# creating a new XGBooster
xgb = XGBooster(
options,
from_model=
'../temp/spam10_data/spam10_data_nbestim_50_maxdepth_3_testsplit_0.2.mod.pkl'
)
# encode it and save the encoding to another file
xgb.encode()
options.explain = [float(v.strip()) for v in s.split(',')]
if tuple(options.explain) in tested:
continue
tested.add(tuple(options.explain))
print(prefix, 'sample {0}: {1}'.format(i, ','.join(s.split(','))))
# calling anchor
timer = resource.getrusage(resource.RUSAGE_CHILDREN).ru_utime + \
resource.getrusage(resource.RUSAGE_SELF).ru_utime
expls = xgb.explain(options.explain)
timer = resource.getrusage(resource.RUSAGE_CHILDREN).ru_utime + \
resource.getrusage(resource.RUSAGE_SELF).ru_utime - timer
times.append(timer)
print(prefix, 'expls:', expls)
print(prefix, 'nof x:', len(expls))
print(prefix, 'timex: {0:.2f}'.format(timer))
print(prefix, 'calls:', xgb.x.calls)
print(prefix, 'Msz x:', max([len(x) for x in expls]))
print(prefix, 'msz x:', min([len(x) for x in expls]))
print(
prefix,
'asz x: {0:.2f}'.format(sum([len(x) for x in expls]) / len(expls)))
print('')
calls.append(xgb.x.calls)
xsize.append(sum([len(x) for x in expls]) / float(len(expls)))
exlen.append(len(expls))
print('')
print('all samples:', len(lines))
# reporting the time spent
print('{0} total time: {1:.2f}'.format(prefix, sum(times)))
print('{0} max time per instance: {1:.2f}'.format(prefix, max(times)))
print('{0} min time per instance: {1:.2f}'.format(prefix, min(times)))
print('{0} avg time per instance: {1:.2f}'.format(prefix,
sum(times) / len(times)))
print('{0} total oracle calls: {1}'.format(prefix, sum(calls)))
print('{0} max oracle calls per instance: {1}'.format(prefix, max(calls)))
print('{0} min oracle calls per instance: {1}'.format(prefix, min(calls)))
print('{0} avg oracle calls per instance: {1:.2f}'.format(
prefix,
float(sum(calls)) / len(calls)))
print('{0} avg number of explanations per instance: {1:.2f}'.format(
prefix,
float(sum(exlen)) / len(exlen)))
print('{0} avg explanation size per instance: {1:.2f}'.format(
prefix,
float(sum(xsize)) / len(xsize)))
print('')
if count > 1:
nof_insts = min(int(count), len(insts))
else:
nof_insts = min(int(len(insts) * count), len(insts))
print(f'considering {nof_insts} instances')
base = os.path.splitext(os.path.basename(data))[0]
mfile = 'temp/{0}/{0}_nbestim_{1}_maxdepth_{2}_testsplit_0.2.mod.pkl'.format(
base, num, adepth)
slog = open(f'results/smt/{base}.log', 'w')
mlog = open(f'results/mx/{base}.log', 'w')
# creating booster objects
sxgb = XGBooster(soptions, from_model=mfile)
sxgb.encode(test_on=None)
mxgb = XGBooster(moptions, from_model=mfile)
mxgb.encode(test_on=None)
stimes = []
mtimes = []
mcalls = []
smem = []
mxmem = []
#with open("/tmp/texture.samples", 'r') as fp:
# insts = [line.strip() for line in fp.readlines()]
for i, inst in enumerate(insts):
if i == nof_insts:
if (options.preprocess_categorical):
preprocess_dataset(options.files[0],
options.preprocess_categorical_files)
exit()
if options.files:
xgb = None
if options.train:
data = Data(filename=options.files[0],
mapfile=options.mapfile,
separator=options.separator,
use_categorical=options.use_categorical)
xgb = XGBooster(options, from_data=data)
train_accuracy, test_accuracy, model = xgb.train()
# read a sample from options.explain
if options.explain:
options.explain = [
float(v.strip()) for v in options.explain.split(',')
]
if options.encode:
if not xgb:
xgb = XGBooster(options, from_model=options.files[0])
# encode it and save the encoding to another file
xgb.encode(test_on=options.explain)