def fit(self):
# Note: the 'ground_truth.csv' file is the test WITH the references that have been removed
# Ask to @teomore to have it
data_directory = self.data_directory.joinpath(self.mode)
self.dict_sub_scores = {}
# This creates a dictonary that holds for each sub the list of the scores (1 for each impression)
for d in self.dfs_subname:
print(f'Getting scores for {d[1]}...')
if os.path.isfile(self.data_directory.joinpath(f'scores/scores_{d[1]}.pkl')): # if the scores were previously computed, simply loads them
with open(self.data_directory.joinpath(f'scores/scores_{d[1]}.pkl'), 'rb') as file:
self.dict_sub_scores[d[1]] = pickle.load(file)
else:
# if there are no scores previously computed...
# first: generate a sub for each impression "column"
# second: score each sub
# third: append each score to the scores' list
scores = []
self.generate_column_subs(d)
for n in tqdm(range(1, 25)):
subm_csv = self.data_directory.joinpath(f'scores/item_{d[1]}_{n}.csv')
mrr = f.score_submissions(subm_csv, self.gt_csv, f.get_reciprocal_ranks)
scores.append(mrr)
self.dict_sub_scores[d[1]] = scores
print('Saving list...')
with open(self.data_directory.joinpath(f'scores/scores_{d[1]}.pkl'), 'wb') as file:
pickle.dump(scores, file)
# Remove files scores/item_{d[1]}_{n}
for n in tqdm(range(1, 25)):
os.remove(self.data_directory.joinpath(f'scores/item_{d[1]}_{n}.csv'))
return self.dict_sub_scores
def __init__(self, sub_name):
self.current_directory = Path(__file__).absolute().parent
self.data_directory = self.current_directory.joinpath('..', 'submissions/hybrid/scores')
self.sub_name = sub_name
self.gt_csv = self.current_directory.joinpath('..', 'submissions/evaluator', 'ground_truth.csv')
self.sub_scores = []
self.subm_csv = self.current_directory.joinpath('..', 'submissions/evaluator', f'{sub_name}.csv')
self.overall_score = f.score_submissions(self.subm_csv, self.gt_csv, f.get_reciprocal_ranks)
def score_sub(self, submission):
#compute the score of a submission using utils/functions.py
mrr = f.score_submissions(submission,
self.gt_csv,
f.get_reciprocal_ranks,
subm_csv_is_file=False)
print(f'Score: {mrr}')
return mrr
def score_sub(self, sub_path, gt_csv, total_score=-1):
cluster_name = os.path.basename(self.data_directory.joinpath(gt_csv))
cluster_name = os.path.splitext(cluster_name)[0]
print("Computing score for"+f"\033[1;35;40m {sub_path}"+ '\033[0;37;40m' +" with cluster" + f"\033[1;35;40m {cluster_name}"+ '\033[0;37;40m')
subm_csv = self.data_directory.joinpath(sub_path)
mrr = f.score_submissions(subm_csv, gt_csv, f.get_reciprocal_ranks)
if total_score != -1:
if mrr > total_score:
print('\033[1;40m Score: '+ f'\033[1;32;40m {mrr}'+ '\033[1;40m :)' + '\033[0;37;40m')
elif mrr < total_score-0.02 and mrr > total_score-0.04:
print('\033[1;40m Score: '+ f'\033[1;31;40m {mrr}'+ '\033[1;40m :(' + '\033[0;37;40m')
elif mrr < total_score - 0.04:
print('\033[1;40m Score: '+ f'\033[1;31;40m {mrr}'+ '\033[1;40m ¯\_(⊙︿⊙)_/¯ <--------- PROBLEM HERE!' + '\033[0;37;40m')
else:
print('\033[1;40m Score: '+ f'\033[1;33;40m {mrr}'+ '\033[1;40m :|' + '\033[0;37;40m')
else:
print('\033[1;40m Score: '+ f'\033[1;32;40m {mrr}'+ '\033[0;37;40m')
return mrr
def get_scores(self):
print(f'Getting scores for {self.sub_name}...')
if os.path.isfile(self.data_directory.joinpath(f'scores_{self.sub_name}.pkl')): # if the scores were previously computed, simply loads them
with open(self.data_directory.joinpath(f'scores_{self.sub_name}.pkl'), 'rb') as file:
self.sub_scores = pickle.load(file)
else:
# if there are no scores previously computed...
# first: generate a sub for each impression "column"
# second: score each sub
# third: append each score to the scores' list
scores = []
self.generate_column_subs()
for n in tqdm(range(1, 25)):
subm_csv = self.data_directory.joinpath(f'item_{self.sub_name}_{n}.csv')
mrr = f.score_submissions(subm_csv, self.gt_csv, f.get_reciprocal_ranks)
scores.append(mrr)
self.sub_scores = scores
print('Saving list...')
with open(self.data_directory.joinpath(f'scores_{self.sub_name}.pkl'), 'wb') as file:
pickle.dump(scores, file)
for n in tqdm(range(1, 25)):
os.remove(self.data_directory.joinpath(f'item_{self.sub_name}_{n}.csv'))