-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtune_ensemble_firstlevel.py
133 lines (100 loc) · 6.18 KB
/
tune_ensemble_firstlevel.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
import os
import optuna
import argparse
import numpy as np
import pandas as pd
import scipy.sparse as sps
import pickle as pkl
from RecSysFramework.DataManager import Dataset
from RecSysFramework.Recommender.DataIO import DataIO
from RecSysFramework.Evaluation.Metrics import ndcg
from RecSysFramework.ExperimentalConfig import EXPERIMENTAL_CONFIG
from RecSysFramework.Utils import load_compressed_csr_matrix, save_compressed_csr_matrix
from xgboost_utils import XGBoostOptimizer, XGBoostSmallOptimizer
from lightgbm_utils import LightGBMOptimizer, LightGBMSmallOptimizer
from catboost_utils import CatboostOptimizer, CatboostSmallOptimizer
from utils import FeatureGenerator, Optimizer, remove_seen, remove_useless_features, get_useless_columns
from utils import create_dataset_from_folder, compress_urm, read_ratings, output_scores, row_minmax_scaling, evaluate, first_level_ensemble, stretch_urm, make_submission
EXPERIMENTAL_CONFIG['n_folds'] = 0
def print_importance(model, n=10):
names = model.feature_name_
importances = model.feature_importances_
order = np.argsort(importances)[::-1]
for i in order[:n]:
print(names[i], importances[i])
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='First level Ensemble hyperparameter optimization')
parser.add_argument('--ntrials', '-t', metavar='TRIALS', type=int, nargs='?', default=200,
help='Number of trials for hyperparameter optimization')
parser.add_argument('--nfolds', '-cv', metavar='FOLDS', type=int, nargs='?', default=5,
help='Number of CV folds for hyperparameter optimization')
parser.add_argument('--force-hpo', '-f', nargs='?', dest="force_hpo", default=False, const=True,
help='Whether to run a new hyperparameter optimization discarding previous ones')
parser.add_argument('--small', '-s', nargs='?', dest="small", default=False, const=True,
help='Whether to train on two different splittings, but on smaller train sets')
parser.add_argument('--gbdt', '-g', metavar='GBDT', type=str, nargs='?', default='lightgbm', dest="gbdt",
choices=['lightgbm', 'xgboost', 'catboost'], help='GBDT model to use')
args = parser.parse_args()
n_trials = args.ntrials
n_folds = args.nfolds
force_hpo = args.force_hpo
if args.gbdt == "xgboost":
if args.small:
optimizer_class = XGBoostSmallOptimizer
else:
optimizer_class = XGBoostOptimizer
elif args.gbdt == "catboost":
if args.small:
optimizer_class = CatboostSmallOptimizer
else:
optimizer_class = CatboostOptimizer
else:
if args.small:
optimizer_class = LightGBMSmallOptimizer
else:
optimizer_class = LightGBMOptimizer
for exam_folder in EXPERIMENTAL_CONFIG['test-datasets']:
exam_train, exam_valid, urm_exam_valid_neg, urm_exam_test_neg = create_dataset_from_folder(exam_folder)
exam_user_mapper, exam_item_mapper = exam_train.get_URM_mapper()
featgen = FeatureGenerator(exam_folder)
urms = featgen.get_urms()
validations = featgen.get_validations()
user_mappers = featgen.get_user_mappers()
item_mappers = featgen.get_item_mappers()
for folder in EXPERIMENTAL_CONFIG['datasets']:
if exam_folder in folder:
for normalization in ["", "-nonorm", "-both"]:
ratings = featgen.load_folder_features(folder)
if normalization in ["", "-both"]:
predictions = featgen.load_algorithms_predictions(folder, normalize=True)
for j in range(len(predictions)):
ratings[j] = ratings[j].merge(predictions[j], on=["user" ,"item"], how="left", sort=True)
if normalization in ["-nonorm", "-both"]:
predictions = featgen.load_algorithms_predictions(folder, normalize=False)
for j in range(len(predictions)):
ratings[j] = ratings[j].merge(predictions[j], on=["user" ,"item"], how="left", sort=True)
useless_cols = get_useless_columns(ratings[-2])
for i in range(len(ratings)):
remove_useless_features(ratings[i], columns_to_remove=useless_cols, inplace=True)
user_factors, item_factors = featgen.load_user_factors(folder, num_factors=12, normalize=True)
for j in range(len(user_factors)):
ratings[j] = ratings[j].merge(item_factors[j], on=["item"], how="left", sort=False)
ratings[j] = ratings[j].merge(user_factors[j], on=["user"], how="left", sort=True)
optimizer = optimizer_class(urms, ratings, validations, n_folds=n_folds)
optimizer.optimize_all(exam_folder, force=force_hpo, n_trials=n_trials, folder=folder, study_name_suffix="-f" + normalization)
results_filename = EXPERIMENTAL_CONFIG['dataset_folder'] + folder + os.sep + \
"{}{}-ensemble-prediction-{}".format(optimizer.NAME, normalization, exam_folder)
er, er_test, result = optimizer.train_cv_best_params(urms[-2], ratings[-2], validations[-1], test_df=ratings[-1])
output_scores(results_filename + "-valid.tsv.gz", er.tocsr(), user_mappers[-2], item_mappers[-2], compress=True)
output_scores(results_filename + "-test.tsv.gz", er_test.tocsr(), user_mappers[-1], item_mappers[-1], compress=True)
print(exam_folder, folder, "Optimization finished:", result)
for fold in range(EXPERIMENTAL_CONFIG['n_folds']):
er, result = optimizer.train_cv_best_params(urms[fold], ratings[fold], validations[fold])
output_scores(results_filename + "-f{}.tsv.gz".format(fold), er.tocsr(),
user_mappers[fold], item_mappers[fold], compress=True)
del optimizer
del validations
del user_mappers
del item_mappers
del urms
del featgen