def recommend(self):
# Compute the indices of the non-target playlists
b = Builder()
nontarget_indices = b.get_nontarget_indices(self.target_tracks)
# Initialize the dataframe
dataframe_list = []
# Apply tfidf on the traspose of URM
URM_T = self.URM.T
URM_tfidf_T = feature_extraction.text.TfidfTransformer().fit_transform(
URM_T)
URM_tfidf = URM_tfidf_T.T
URM_tfidf_csr = URM_tfidf.tocsr()
# Weighted average of S_ICM and S_UCM
S_avg = (self.avg * self.S_ICM) + ((1 - self.avg) * self.S_UCM)
ordered_ps_id = b.get_ordered_playlists_id()
print('Predicting...', flush=True)
for i in tqdm(range(0, self.num_playlist_to_recommend)):
# Iterate over indices of target playlists
index = b.get_target_playlist_index(self.target_playlists[i])
# Compute the indices of the known tracks
known_indices = np.nonzero(self.URM[index].toarray().flatten())
# Consider order for ordered playlists
if index in ordered_ps_id:
URM_tfidf_adjusted = b.adjust_order(URM_tfidf_csr[index, :],
index)
else:
URM_tfidf_adjusted = URM_tfidf_csr[index, :]
# Calculate recommenders contributions
avg_prediction = URM_tfidf_adjusted * S_avg
slimBPR_prediction = URM_tfidf_adjusted * self.Slim
# Weighted sum of recommendations
URM_row = (self.alfa * avg_prediction) + (
(1 - self.alfa) * slimBPR_prediction)
# Make top-10 prediction
URM_row_flatten = URM_row.toarray().flatten()
top_10_indices = b.get_top_10_indices(URM_row_flatten,
nontarget_indices,
known_indices, [])
top_10_tracks = b.get_top_10_tracks_from_indices(top_10_indices)
top_10_tracks_string = ' '.join([str(i) for i in top_10_tracks])
# Create dataset
if self.is_test:
dataframe_list.append(
[self.target_playlists[i], top_10_tracks])
else:
dataframe_list.append(
[self.target_playlists[i], top_10_tracks_string])
dataframe = pd.DataFrame(dataframe_list,
columns=['playlist_id', 'track_ids'])
return dataframe
