def rf_model(read_csv=True):
print 'rf_model'
train_df, test_df = get_train_test_data(cache=read_csv, include_sessions=True)
cols = [i for i in train_df.columns if i not in EXCLUDE_COLS]
X = train_df[cols]
y = train_df['country_destination']
clf = RandomForestClassifier(n_estimators=20)
# make submission
# clf.fit(X.values, y)
# py.test.set_trace()
# test_data = test_df[[i for i in test_df.columns if i not in EXCLUDE_COLS]]
# y_pred = bst.predict_proba(test_data)
# print 'predicted submission prob'
# py.test.set_trace()
# sub = create_kaggle_submission(y_pred, test_df['id'], None)
# print 'created kaggle submission'
# kf = KFold(len(X), n_folds=10, random_state=42)
# score = cross_val_score(bst, X, y, cv=kf, scoring=ndcg_scorer)
param_dist = {"max_depth": [3, None],
"max_features": sp_randint(1, 11),
"min_samples_split": sp_randint(1, 11),
"min_samples_leaf": sp_randint(1, 11),
"bootstrap": [True, False]
}
n_iter_search = 20
random_search = RandomizedSearchCV(clf, param_distributions=param_dist,
n_iter=n_iter_search, scoring=ndcg_scorer)
start = time()
random_search.fit(X, y)
print("RandomizedSearchCV took %.2f seconds for %d candidates"
" parameter settings." % ((time() - start), n_iter_search))
report(random_search.grid_scores_)
py.test.set_trace()
return
def xgb_model(read_csv=True):
print 'xgb_model randomcv yr > 2013'
train_df, test_df = get_train_test_data(cache=read_csv, include_sessions=False)
train_df = train_df[train_df['tfa_year'] > 2013]
cols = [i for i in train_df.columns if i not in EXCLUDE_COLS]
X = train_df[cols]
y = train_df['country_destination']
#start classifier
bst = xgb.XGBClassifier(nthread=4)
# bst = xgb.XGBClassifier(max_depth=2, nthread=4,
# n_estimators=50,subsample=0.4,learning_rate=0.0.05)
train_X, test_X, train_y, test_y = train_test_split(X, y, test_size=10000)
# bst.fit(test_X, test_y)
param_dist = {"max_depth": [2,4,6],
"learning_rate": [0.05, 0.1, 0.15, 0.2],
"n_estimators": [30, 50, 70],
# "min_samples_leaf": sp_randint(1, 11),
# "min_samples_split": sp_randint(1,11),
'subsample': [0.4, 0.5, 0.6]
# 'max_features': [20, 50, 100]
}
n_iter_search = 20
random_search = RandomizedSearchCV(bst, param_distributions=param_dist,
n_iter=n_iter_search, scoring=ndcg_scorer)
start = time()
random_search.fit(test_X, test_y)
print("RandomizedSearchCV took %.2f seconds for %d candidates"
" parameter settings." % ((time() - start), n_iter_search))
report(random_search.grid_scores_)
# bagging
# clfbag = BaggingClassifier(bst, n_estimators=5, max_samples=5000)
# clfbag.fit(train_X, train_y)
# y_pred = clfbag.predict_proba(test_X)
# print 'predicted prob'
# score = ndcg_score(test_y, y_pred)
# apply learned model
# bst.fit(X.values, y)
# py.test.set_trace()
test_data = test_df[[i for i in test_df.columns if i not in EXCLUDE_COLS]]
y_pred = random_search.predict_proba(test_data)
py.test.set_trace()
# kaggle_test = pd.read_csv('test.csv')
sub = create_kaggle_submission(y_pred, test_df['id'], 0.841)
py.test.set_trace()
# print 'created kaggle sub'
# kf = KFold(len(X), n_folds=10, random_state=42)
# score = cross_val_score(bst, X, y, cv=kf, scoring=ndcg_scorer)
# param_dist = {"max_depth": [3, None],
# "max_features": sp_randint(1, 11),
# "min_samples_split": sp_randint(1, 11),
# "min_samples_leaf": sp_randint(1, 11),
# "bootstrap": [True, False],
# "criterion": ["gini", "entropy"]}
# n_iter_search = 20
# random_search = RandomizedSearchCV(clf, param_distributions=param_dist,
# n_iter=n_iter_search, scoring=ndcg_scorer)
# start = time()
# random_search.fit(X, y)
# print("RandomizedSearchCV took %.2f seconds for %d candidates"
# " parameter settings." % ((time() - start), n_iter_search))
# report(random_search.grid_scores_)
# py.test.set_trace()
# sub = create_kaggle_submission(y_pred, test_df['id'], np.mean(score))
# py.test.set_trace()
#end classifier
""" trying cross valid
