num_cv=1, num_folds=10, scale='log')
roc_aucs = pd.DataFrame(np.array(roc_aucs), index=['auc', 'std'],
columns = map(lambda x: x['model'], params_ens)+['ENS'])
roc_aucs.to_csv('data/submi/roc_aucs.csv', float_format='%11.6f')
feature_set.to_csv('data/submi/features.csv')
print "cross validation results:"
print roc_aucs
############################################################################
# fit and predict
############################################################################
result = fb_funcs.fit_and_predict(info_humans, info_bots, info_test,
params=params_ens, scale='log')
y_test_proba = result['y_test_proba']
ytps = result['ytps']
# feature_importances = pd.DataFrame(np.array([result['features'],
# result['importances']]).T)
############################################################################
# submission file generation
############################################################################
submissionfile = 'data/submi/sub_ens.csv'
testfile = 'data/test.csv'
print "writing a submission file..."
write_submission(y_test_proba, info_test.index, testfile, submissionfile)
roc_auc = np.array(roc_auc)
roc_auc_std = np.array(roc_auc_std)
clf_score = np.array(clf_score)
print ""
print roc_auc.mean(), roc_auc_std.mean()
print clf_score.mean(), clf_score.std()
# print tpr_50
############################################################################
# fit and predict
############################################################################
y_test_proba, y_train_proba, _\
= fit_and_predict(info_humans, info_bots, info_test, model='ET',
n_estimators=1000, p_use=None, plotting=True)
############################################################################
# xgboost: CV
############################################################################
# y_pred, ytrain_pred, cv_result \
# = fit_and_predict(info_humans, info_bots, info_test,
# n_estimators=20, p_use=None, cv=5)
# auc = []
# for i in range(len(cv_result)):
# auc.append(float(cv_result[i].split('\t')[1].split(':')[1].split('+')[0]))
# best_itr = np.argmax(auc)
# auc_std = float(cv_result[11].split('\t')[1].split('+')[1])
.union(dinfo_bots.keys())\
.union(dinfo_test.keys())
info_humans = append_device(info_humans, dinfo_humans, devices_appended)
info_bots = append_device(info_bots, dinfo_bots, devices_appended)
info_test = append_device(info_test, dinfo_test, devices_appended)
info_humans.fillna(0, inplace=True)
info_bots.fillna(0, inplace=True)
info_test.fillna(0, inplace=True)
info_humans.drop('merchandise', inplace=True, axis=1)
info_bots.drop('merchandise', inplace=True, axis=1)
info_test.drop('merchandise', inplace=True, axis=1)
y_test_proba, y_train_proba, _, features\
= fit_and_predict(info_humans, info_bots, info_test, model='ET',
params=params)
device_importance = ['phone46', 'num_bids', 'phone143',
'num_ips', 'num_aucs', 'phone55',
'num_urls', 'phone63', 'phone2287',
'phone2330', 'phone239', 'phone110',
'phone3359', 'phone168', 'num_devices',
'phone22', 'num_countries', 'phone33',
'phone205', 'phone150', 'phone1026',
'phone728', 'phone136', 'phone25',
'phone224', 'phone640', 'phone1166',
'phone892', 'phone2955', 'phone1013',
'phone195', 'phone58', 'phone4479',
'phone469', 'phone90', 'phone15',
'phone996', 'phone5479', 'phone792',
'phone4']
