-
Notifications
You must be signed in to change notification settings - Fork 248
/
run_IJCAI_17_DMF.py
338 lines (248 loc) · 14.6 KB
/
run_IJCAI_17_DMF.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
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on 28/04/19
@author: Simone Boglio
"""
import numpy as np
import os, traceback, argparse
from functools import partial
from Base.Evaluation.Evaluator import EvaluatorNegativeItemSample
from Recommender_import_list import *
# from Conferences.IJCAI.DMF_our_interface.DMFWrapper import DMF_BCE_RecommenderWrapper, DMF_NCE_RecommenderWrapper
from ParameterTuning.run_parameter_search import runParameterSearch_Collaborative
from ParameterTuning.SearchSingleCase import SearchSingleCase
from ParameterTuning.SearchAbstractClass import SearchInputRecommenderArgs
from Utils.ResultFolderLoader import ResultFolderLoader, generate_latex_hyperparameters
from Utils.assertions_on_data_for_experiments import assert_implicit_data, assert_disjoint_matrices
from Utils.plot_popularity import plot_popularity_bias, save_popularity_statistics
from Conferences.IJCAI.DMF_our_interface.Movielens1MReader.Movielens1MReader import Movielens1MReader
from Conferences.IJCAI.DMF_our_interface.Movielens100KReader.Movielens100KReader import Movielens100KReader
from Conferences.IJCAI.DMF_our_interface.AmazonMusicReader.AmazonMusicReader import AmazonMusicReader
from Conferences.IJCAI.DMF_our_interface.AmazonMovieReader.AmazonMovieReader import AmazonMovieReader
def cold_items_statistics(URM_train, URM_validation, URM_test, URM_test_negative):
# Cold items experiment
import scipy.sparse as sps
URM_train_validation = URM_train + URM_validation
n_users, n_items = URM_train_validation.shape
item_in_train_flag = np.ediff1d(sps.csc_matrix(URM_train_validation).indptr) > 0
item_in_test_flag = np.ediff1d(sps.csc_matrix(URM_test).indptr) > 0
test_item_not_in_train_flag = np.logical_and(item_in_test_flag, np.logical_not(item_in_train_flag))
test_item_in_train_flag = np.logical_and(item_in_test_flag, item_in_train_flag)
print("The test data contains {} unique items, {} ({:.2f} %) of them never appear in train data".format(
item_in_test_flag.sum(),
test_item_not_in_train_flag.sum(),
test_item_not_in_train_flag.sum()/item_in_test_flag.sum()*100,
))
def read_data_split_and_search(dataset_name,
flag_baselines_tune = False,
flag_DL_article_default = False, flag_DL_tune = False,
flag_print_results = False):
result_folder_path = "result_experiments/{}/{}_{}/".format(CONFERENCE_NAME, ALGORITHM_NAME, dataset_name)
if dataset_name == 'amazon_music_original':
dataset = AmazonMusicReader(result_folder_path, original = True)
elif dataset_name == 'amazon_music_ours':
dataset = AmazonMusicReader(result_folder_path, original = False)
elif dataset_name == 'amazon_movie':
dataset = AmazonMovieReader(result_folder_path)
elif dataset_name == 'movielens100k':
dataset = Movielens100KReader(result_folder_path)
elif dataset_name == 'movielens1m':
dataset = Movielens1MReader(result_folder_path)
else:
print("Dataset name not supported, current is {}".format(dataset_name))
return
print ('Current dataset is: {}'.format(dataset_name))
URM_train = dataset.URM_DICT["URM_train"].copy()
URM_validation = dataset.URM_DICT["URM_validation"].copy()
URM_test = dataset.URM_DICT["URM_test"].copy()
URM_test_negative = dataset.URM_DICT["URM_test_negative"].copy()
# Ensure DISJOINT sets. Do not ensure IMPLICIT data because the algorithm needs explicit data
assert_disjoint_matrices([URM_train, URM_validation, URM_test, URM_test_negative])
cold_items_statistics(URM_train, URM_validation, URM_test, URM_test_negative)
# If directory does not exist, create
if not os.path.exists(result_folder_path):
os.makedirs(result_folder_path)
algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
plot_popularity_bias([URM_train + URM_validation, URM_test],
["Training data", "Test data"],
result_folder_path + algorithm_dataset_string + "popularity_plot")
save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
["Full data", "Training data", "Test data"],
result_folder_path + algorithm_dataset_string + "popularity_statistics")
collaborative_algorithm_list = [
Random,
TopPop,
UserKNNCFRecommender,
ItemKNNCFRecommender,
P3alphaRecommender,
RP3betaRecommender,
PureSVDRecommender,
NMFRecommender,
IALSRecommender,
MatrixFactorization_BPR_Cython,
MatrixFactorization_FunkSVD_Cython,
EASE_R_Recommender,
SLIM_BPR_Cython,
SLIMElasticNetRecommender,
]
metric_to_optimize = "NDCG"
n_cases = 50
n_random_starts = 15
cutoff_list_validation = [10]
cutoff_list_test = [5, 10, 20]
evaluator_validation = EvaluatorNegativeItemSample(URM_validation, URM_test_negative, cutoff_list=cutoff_list_validation)
evaluator_test = EvaluatorNegativeItemSample(URM_test, URM_test_negative, cutoff_list=cutoff_list_test)
runParameterSearch_Collaborative_partial = partial(runParameterSearch_Collaborative,
URM_train = URM_train,
URM_train_last_test = URM_train + URM_validation,
metric_to_optimize = metric_to_optimize,
evaluator_validation_earlystopping = evaluator_validation,
evaluator_validation = evaluator_validation,
evaluator_test = evaluator_test,
output_folder_path = result_folder_path,
parallelizeKNN = False,
allow_weighting = True,
resume_from_saved = True,
n_cases = n_cases,
n_random_starts = n_random_starts)
if flag_baselines_tune:
for recommender_class in collaborative_algorithm_list:
try:
runParameterSearch_Collaborative_partial(recommender_class)
except Exception as e:
print("On recommender {} Exception {}".format(recommender_class, str(e)))
traceback.print_exc()
################################################################################################
######
###### DL ALGORITHM
######
"""
NOTICE: We did not upload the source code of DMF as it was not publicly available and the original
authors did not respond to our request to add it to this repository
"""
if flag_DL_article_default:
if dataset_name in ['amazon_music_original', 'amazon_music_ours']:
last_layer_size = 128
else:
last_layer_size = 64
article_hyperparameters = {'epochs': 300,
'learning_rate': 0.0001,
'batch_size': 256,
'num_negatives': 7, # As reported in the "Detailed implementation" section of the original paper
'last_layer_size': last_layer_size,
}
earlystopping_hyperparameters = {'validation_every_n': 5,
'stop_on_validation': True,
'lower_validations_allowed': 5,
'evaluator_object': evaluator_validation,
'validation_metric': metric_to_optimize,
}
#
# try:
#
#
# parameterSearch = SearchSingleCase(DMF_NCE_RecommenderWrapper,
# evaluator_validation=evaluator_validation,
# evaluator_test=evaluator_test)
#
# recommender_input_args = SearchInputRecommenderArgs(
# CONSTRUCTOR_POSITIONAL_ARGS = [URM_train],
# FIT_KEYWORD_ARGS = earlystopping_hyperparameters)
#
# recommender_input_args_last_test = recommender_input_args.copy()
# recommender_input_args_last_test.CONSTRUCTOR_POSITIONAL_ARGS[0] = URM_train + URM_validation
#
# parameterSearch.search(recommender_input_args,
# recommender_input_args_last_test = recommender_input_args_last_test,
# fit_hyperparameters_values = article_hyperparameters,
# output_folder_path = result_folder_path,
# resume_from_saved = True,
# output_file_name_root = DMF_NCE_RecommenderWrapper.RECOMMENDER_NAME)
#
#
#
# except Exception as e:
#
# print("On recommender {} Exception {}".format(DMF_NCE_RecommenderWrapper, str(e)))
# traceback.print_exc()
#
#
#
# try:
#
#
# parameterSearch = SearchSingleCase(DMF_BCE_RecommenderWrapper,
# evaluator_validation=evaluator_validation,
# evaluator_test=evaluator_test)
#
# recommender_input_args = SearchInputRecommenderArgs(
# CONSTRUCTOR_POSITIONAL_ARGS = [URM_train],
# FIT_KEYWORD_ARGS = earlystopping_hyperparameters)
#
# recommender_input_args_last_test = recommender_input_args.copy()
# recommender_input_args_last_test.CONSTRUCTOR_POSITIONAL_ARGS[0] = URM_train + URM_validation
#
# parameterSearch.search(recommender_input_args,
# recommender_input_args_last_test = recommender_input_args_last_test,
# fit_hyperparameters_values = article_hyperparameters,
# output_folder_path = result_folder_path,
# resume_from_saved = True,
# output_file_name_root = DMF_BCE_RecommenderWrapper.RECOMMENDER_NAME)
#
#
# except Exception as e:
#
# print("On recommender {} Exception {}".format(DMF_BCE_RecommenderWrapper, str(e)))
# traceback.print_exc()
################################################################################################
######
###### PRINT RESULTS
######
if flag_print_results:
n_test_users = np.sum(np.ediff1d(URM_test.indptr)>=1)
file_name = "{}..//{}_{}_".format(result_folder_path, ALGORITHM_NAME, dataset_name)
result_loader = ResultFolderLoader(result_folder_path,
base_algorithm_list = None,
other_algorithm_list = [DMF_NCE_RecommenderWrapper, DMF_BCE_RecommenderWrapper],
KNN_similarity_list = KNN_similarity_to_report_list,
ICM_names_list = None,
UCM_names_list = None)
result_loader.generate_latex_results(file_name + "{}_latex_results.txt".format("article_metrics"),
metrics_list = ["HIT_RATE", "NDCG"],
cutoffs_list = cutoff_list_validation,
table_title = None,
highlight_best = True)
result_loader.generate_latex_results(file_name + "{}_latex_results.txt".format("all_metrics"),
metrics_list = ["PRECISION", "RECALL", "MAP_MIN_DEN", "MRR", "NDCG", "F1", "HIT_RATE", "ARHR_ALL_HITS",
"NOVELTY", "DIVERSITY_MEAN_INTER_LIST", "DIVERSITY_HERFINDAHL", "COVERAGE_ITEM", "DIVERSITY_GINI", "SHANNON_ENTROPY"],
cutoffs_list = [10],
table_title = None,
highlight_best = True)
result_loader.generate_latex_time_statistics(file_name + "{}_latex_results.txt".format("time"),
n_evaluation_users=n_test_users,
table_title = None)
if __name__ == '__main__':
ALGORITHM_NAME = "DMF"
CONFERENCE_NAME = "IJCAI"
parser = argparse.ArgumentParser()
parser.add_argument('-b', '--baseline_tune', help="Baseline hyperparameter search", type = bool, default = False)
parser.add_argument('-a', '--DL_article_default', help="Train the DL model with article hyperparameters", type = bool, default = False)
parser.add_argument('-p', '--print_results', help="Print results", type = bool, default = True)
input_flags = parser.parse_args()
print(input_flags)
KNN_similarity_to_report_list = ["cosine", "dice", "jaccard", "asymmetric", "tversky"]
dataset_list = ['amazon_music_original', 'amazon_music_ours', 'movielens100k', 'amazon_movie', 'movielens1m']
for dataset_name in dataset_list:
read_data_split_and_search(dataset_name,
flag_baselines_tune=input_flags.baseline_tune,
flag_DL_article_default= input_flags.DL_article_default,
flag_print_results = input_flags.print_results,
)
if input_flags.print_results:
generate_latex_hyperparameters(result_folder_path="result_experiments/{}/".format(CONFERENCE_NAME),
algorithm_name=ALGORITHM_NAME,
experiment_subfolder_list=dataset_list,
other_algorithm_list= [DMF_NCE_RecommenderWrapper, DMF_BCE_RecommenderWrapper],
KNN_similarity_to_report_list = KNN_similarity_to_report_list,
split_per_algorithm_type = True)