if len(sys.argv) <= 1:
print "so what pkl do you want me to read, hm?"
sys.exit()
randomForest = ensemble.RandomForestClassifier(
verbose=False, n_estimators=80, min_samples_split=10, max_depth=14, bootstrap=False, n_jobs=16
)
randomForestRegress = ensemble.RandomForestClassifier(
verbose=True, n_estimators=80, min_samples_split=10, max_depth=14, n_jobs=15
)
gradBoost = ensemble.GradientBoostingClassifier(verbose=True, n_estimators=100, min_samples_split=10, max_depth=7)
with open(sys.argv[1]) as infile:
train, _ = pickle.load(infile)
if type(train[1][0]) == list:
ids, info, labels = kddutil.notrash(*train)
info = kddutil.bound(info, max=10000, min=-10000)
else:
ids, info, labels = train
print "assuming compacted data; skip preprocessing"
print "Random Forest"
print kddutil.evaluate_k(randomForest, ids, info, labels, fold=3) # , postprocess=kddutil.disambiguate)
# print "Gradient Boosting"
# print kddutil.evaluate(gradBoost, ids, info, labels)
- also removes duplicates, and infinities
'''
import kddutil
import cPickle as pickle
import numpy as np
import sys
if len(sys.argv) <= 2:
print "pickelino pickelouto"
sys.exit()
print "reading"
with open(sys.argv[1],"rb") as infile:
train, test = pickle.load(infile)
print "numpyifying"
id, feat, lab = kddutil.notrash(*train)
feat = kddutil.bound(feat, min=-10000, max=+10000)
train = np.array(id, np.int32), np.array(feat, np.float16), np.array(lab, np.bool)
id, feat = kddutil.notrash(*test)
feat = kddutil.bound(feat, min=-10000, max=+10000)
test = np.array(id, np.int32), np.array(feat, np.float16)
print "writing"
with open(sys.argv[2],"wb") as outfile:
pickle.dump((train,test), outfile)
gradBoost = ensemble.GradientBoostingClassifier(verbose=True
, n_estimators=100
, min_samples_split=10
, max_depth=5
)
classifier = randomForest
with open(sys.argv[1]) as infile:
train, test = pickle.load(infile)
if type(train[1][0]) == list:
train_ids, train_set, labels = kddutil.notrash(*train)
test_ids, test_set = kddutil.notrash(*test)
train_set = kddutil.bound(train_set, max=10000, min=-10000)
test_set = kddutil.bound(test_set, max=10000, min=-10000)
else:
train_ids, train_set, labels = train
test_ids, test_set = test
print "assuming compacted data; skip preprocessing"
classifier.fit(train_set, labels)
predictions = classifier.predict_proba(test_set)[:,1]
print "writing to output.csv"
kddutil.write_csv(test_ids, predictions)
