def runTest(self):
base_dir = os.path.dirname(os.path.realpath(__file__))
with open(
os.path.join(os.path.dirname(base_dir),
'config/recommender.json')) as data_file:
json_config = json.load(data_file)
rec_system = RecommenderSystem()
self.assertEqual(rec_system.hyperparameters,
json_config['recommender']['hyperparameters'])
self.assertEqual(rec_system.config.config_dict,
json_config['recommender'])
n_factors = self.n_factors
rec_system.initializer.config['n_factors'] = n_factors
rec_system.content_based.n_factors = n_factors
rec_system.content_based.hyperparameters['n_factors'] = n_factors
rec_system.collaborative_filtering.n_factors = n_factors
rec_system.collaborative_filtering.hyperparameters[
'n_factors'] = n_factors
self.assertTrue(isinstance(rec_system.evaluator, Evaluator))
self.assertTrue(isinstance(rec_system.content_based, ContentBased))
self.assertTrue(
isinstance(rec_system.collaborative_filtering,
CollaborativeFiltering))
self.assertTrue(
isinstance(rec_system.content_based, AbstractRecommender))
if rec_system.config.config_dict['recommender'] == 'userbased':
self.assertTrue(
isinstance(rec_system.recommender, CollaborativeFiltering))
if rec_system.config.config_dict['recommender'] == 'itembased':
self.assertTrue(isinstance(rec_system.recommender, ContentBased))
self.assertEqual(rec_system.content_based.n_items, self.documents)
self.assertEqual(rec_system.content_based.n_factors, n_factors)
rec_system.train()
def run_experiment(self):
"""
Runs experiment
"""
all_results = [[
'n_factors', '_lambda', 'desc', 'rmse', 'train_recall',
'test_recall', 'recall_at_200', 'ratio', 'mrr @ 5', 'ndcg @ 5',
'mrr @ 10', 'ndcg @ 10'
]]
runs = RunsLoader()
for run_idx, config_dict in enumerate(
runs.get_runnable_recommenders()):
if run_idx:
print(
"\n___________________________________________________________________________________________"
)
this_config = config_dict.copy()
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=this_config,
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more,
random_seed=self.random_seed)
print(
"Run #%d %s: " %
((run_idx + 1), recommender.config.get_description()),
recommender.content_based, recommender.collaborative_filtering,
", with: ", recommender.config.config_dict)
results = recommender.train()
report_str = 'Summary: Test sum {:.2f}, Train sum {:.2f}, Final error {:.5f}, train recall {:.5f}, '\
'test recall {:.5f}, recall@200 {:.5f}, '\
'ratio {:.5f}, mrr@5 {:.5f}, '\
'ndcg@5 {:.5f}, mrr@10 {:.5f}, ndcg@10 {:.5f}'
print(report_str.format(*results))
current_result = [
recommender.hyperparameters['n_factors'],
recommender.hyperparameters['_lambda'],
recommender.config.get_description()
]
current_result.extend(results)
all_results.append(current_result)
GridSearch(recommender, {},
self.verbose,
report_name='experiment_results').dump_csv(all_results)
def run_recommender(self):
"""
Runs recommender
"""
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more,
random_seed=self.random_seed)
results = recommender.train()
report_str = 'Summary: Test sum {:.2f}, Train sum {:.2f}, Final error {:.5f}, train recall {:.5f}, '\
'test recall {:.5f}, recall@200 {:.5f}, '\
'ratio {:.5f}, mrr@5 {:.5f}, '\
'ndcg@5 {:.5f}, mrr@10 {:.5f}, ndcg@10 {:.5f}'
print(report_str.format(*results))
recommender.dump_recommendations(200)
def run_recommender(self):
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more)
recommender.train()
recommender.get_evaluation_report()
def run_grid_search(self):
"""
Runs grid search
"""
hyperparameters = {
'_lambda': [0.01],
'n_factors': [50, 100, 150, 200, 250, 300]
}
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more,
random_seed=self.random_seed)
GS = GridSearch(recommender, hyperparameters, self.verbose)
best_params, all_results = GS.train()
def run_experiment_with_gridsearch(self):
"""
Runs experiment after running grid search.
"""
print("Getting Userbased hyperparameters")
userbased_configs = {
'_lambda': [0.01],
'n_factors': [50, 100, 150, 200, 250, 300]
}
self.config.set_recommender_type('userbased')
self.config.set_iterations(5)
self.config.set_folds_num(1)
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=self.config.get_all_config(),
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=False,
train_more=self.train_more,
random_seed=self.random_seed)
userbased_hyperparameters, userbased_gridsearch_results =\
GridSearch(recommender, userbased_configs, self.verbose, report_name='grid_search_userbased').train()
print("Userbased hyperparameters:", userbased_hyperparameters)
print("Getting Itembased hyperparameters")
itembased_configs = {
'_lambda': [0.01],
'n_factors': [50, 100, 150, 200, 250, 300]
}
self.config.set_recommender_type('itembased')
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=self.config.get_all_config(),
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=False,
train_more=self.train_more,
random_seed=self.random_seed)
itembased_hyperparameters, itembased_gridsearch_results =\
GridSearch(recommender, itembased_configs, self.verbose, report_name='grid_search_itembased').train()
print("Itembased hyperparameters:", itembased_hyperparameters)
for _ in range(5):
print('.')
print('Grid search done...')
print('')
print("Userbased hyperparameters:", userbased_hyperparameters)
print("Itembased hyperparameters:", itembased_hyperparameters)
all_results = [[
'n_factors', '_lambda', 'desc', 'rmse', 'train_recall',
'test_recall', 'recall_at_200', 'ratio', 'mrr @ 5', 'ndcg @ 5',
'mrr @ 10', 'ndcg @ 10'
]]
runs = RunsLoader()
for run_idx, config_dict in enumerate(
runs.get_runnable_recommenders()):
if run_idx:
print(
"\n___________________________________________________________________________________________"
)
this_config = config_dict.copy()
if this_config['recommender']['recommender'] == 'itembased':
if itembased_hyperparameters:
this_config['recommender'][
'hyperparameters'] = itembased_hyperparameters.copy()
elif this_config['recommender']['recommender'] == 'userbased':
if userbased_hyperparameters:
this_config['recommender'][
'hyperparameters'] = userbased_hyperparameters.copy()
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=this_config,
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more,
random_seed=self.random_seed)
print(
"Run #%d %s: " %
((run_idx + 1), recommender.config.get_description()),
recommender.content_based, recommender.collaborative_filtering,
", with: ", recommender.config.config_dict)
results = recommender.train()
report_str = 'Summary: Test sum {:.2f}, Train sum {:.2f}, Final error {:.5f}, train recall {:.5f}, '\
'test recall {:.5f}, recall@200 {:.5f}, '\
'ratio {:.5f}, mrr@5 {:.5f}, '\
'ndcg@5 {:.5f}, mrr@10 {:.5f}, ndcg@10 {:.5f}'
print(report_str.format(*results))
current_result = [
recommender.hyperparameters['n_factors'],
recommender.hyperparameters['_lambda'],
recommender.config.get_description()
]
current_result.extend(results)
all_results.append(current_result)
GridSearch(recommender, {},
self.verbose,
report_name='experiment_results').dump_csv(all_results)
def run_experiment(self):
print("Getting Userbased hyperparameters")
userbased_configs = {
'_lambda': [0.00001, 0.01, 0.1, 0.5, 10],
'n_factors': [100, 200, 300, 400, 500]
}
self.config.set_recommender_type('userbased')
print(self.config.config_dict)
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=self.config.get_all_config(),
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more)
userbased_hyperparameters, userbased_gridsearch_results =\
GridSearch(recommender, userbased_configs, self.verbose, report_name='grid_search_userbased').train()
print("Userbased hyperparameters:", userbased_hyperparameters)
print("Getting Itembased hyperparameters")
itembased_configs = {
'_lambda': [0.00001],
'n_factors': [500, 400, 300, 200, 100]
}
self.config.set_recommender_type('itembased')
print(self.config.config_dict)
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=self.config.get_all_config(),
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more)
itembased_hyperparameters, itembased_gridsearch_results =\
GridSearch(recommender, itembased_configs, self.verbose, report_name='grid_search_itembased').train()
print("Itembased hyperparameters:", itembased_hyperparameters)
all_results = [[
'n_factors', '_lambda', 'rmse', 'train_recall', 'test_recall',
'recall_at_200', 'ratio', 'mrr @ 5', 'ndcg @ 5', 'mrr @ 10',
'ndcg @ 10'
]]
runs = RunsLoader()
for config_dict in runs.get_runnable_recommenders():
this_config = config_dict.copy()
if this_config['recommender']['recommender'] == 'itembased':
if itembased_hyperparameters:
this_config['recommender'][
'hyperparameters'] = itembased_hyperparameters.copy()
elif this_config['recommender']['recommender'] == 'userbased':
if userbased_hyperparameters:
this_config['recommender'][
'hyperparameters'] = userbased_hyperparameters.copy()
recommender = RecommenderSystem(
abstracts_preprocessor=self.abstracts_preprocessor,
ratings=self.ratings,
config=this_config,
verbose=self.verbose,
load_matrices=self.load_matrices,
dump_matrices=self.dump,
train_more=self.train_more)
print("Running: ", recommender.content_based,
recommender.collaborative_filtering, ", with: ",
recommender.config.config_dict)
recommender.train()
current_result = [
recommender.hyperparameters['n_factors'],
recommender.hyperparameters['_lambda']
]
current_result.extend(recommender.get_evaluation_report())
all_results.append(current_result)
GridSearch(recommender, {},
self.verbose,
report_name='experiment_results').dump_csv(all_results)