class ListHybridRecommender():
def __init__(self,
urm,
urm_t,
icm,
icm2,
enable_dict,
urm_test,
recalcSLIM=True):
self.urm = urm
self.setEnables(enable_dict)
self.item_item = IIHybridRecommender(urm, icm, icm2)
self.item_item.fit(item_weight=0.4, cbf1_weight=0.25, cbf2_weight=0.1)
if self.enableUSER:
self.cbu = CollaborativeFiltering()
self.cbu.fit(urm_t, k=100, h=0, mode='user')
if self.enableRP3B:
self.rp3b = RP3betaRecommender(urm.getCSR())
if self.enableP3A:
self.p3a = P3alpha(urm.getCSR())
self.p3a.fit(topK=80,
alpha=1,
min_rating=0,
implicit=True,
normalize_similarity=True)
if self.enableSLIM:
if recalcSLIM:
choice = 2
logFile = open("SLIM_BPR_Cython.txt", "a")
self.slim = SLIM_BPR_Cython(urm.getCSR(),
recompile_cython=False,
positive_threshold=0,
URM_validation=urm_test.getCSR(),
final_model_sparse_weights=True,
train_with_sparse_weights=False)
self.slim.fit(epochs=100,
validation_every_n=1,
logFile=logFile,
batch_size=5,
topK=200,
sgd_mode="adagrad",
learning_rate=0.075)
self.slim_sim = self.slim.get_similarity()
# with open('slim_sub.pkl', 'wb') as output:
# pickle.dump(self.slim, output, pickle.HIGHEST_PROTOCOL)
else:
with open('slim_test.pkl', 'rb') as input:
self.slim = pickle.load(input)
def fit(self, weights_dict=None, norm="none", w_method="count"):
self.norm_method = norm
self.weights_dict = weights_dict
self.w_method = w_method
self.item_item_weight = weights_dict.get('item_item_weight', 0)
self.rp3b_weight = weights_dict.get('rp3b_weight', 0)
self.slim_weight = weights_dict.get('slim_weight', 0)
self.user_weight = weights_dict.get('user_weight', 0)
self.p3a_weight = weights_dict.get('p3a_weight', 0)
def s_recommend(self, user, nRec=10):
weighting_dict = {}
#recommended_items_item_item = self.normalize_row(self.item_item.get_pred_row(user), method=self.norm_method)
recommended_items_item_item = self.item_item.s_recommend(
user, nRec).tolist()
weighting_dict['ii'] = (recommended_items_item_item,
self.item_item_weight)
recommended_items_rp3b = None
if (self.enableSVD):
#recommended_items_rp3b = self.normalize_row(self.svd.get_pred_row(user), method=self.norm_method)
recommended_items_rp3b = self.rp3b.s_recommend(user, nRec).tolist()
weighting_dict['rp3b'] = (recommended_items_rp3b, self.rp3b_weight)
recommended_items_p3a = None
if (self.enableP3A):
# recommended_items_svd = self.normalize_row(self.svd.get_pred_row(user), method=self.norm_method)
recommended_items_p3a = self.p3a.s_recommend(user, nRec)
weighting_dict['p3a'] = (recommended_items_p3a, self.p3a_weight)
recommended_items_user = None
if (self.enableUSER):
recommended_items_user = self.cbu.s_recommend(user, nRec).tolist()
weighting_dict['user'] = (recommended_items_user, self.user_weight)
recommended_items_slim = None
if (self.enableSLIM):
#recommended_items_slim = self.normalize_row(self.getSlimRow(user), method=self.norm_method)
recommended_items_slim = self.slim.s_recommend(user, nRec)
weighting_dict['slim'] = (recommended_items_slim, self.slim_weight)
return self.list_weighter(weighting_dict, nRec, 0, self.w_method)
#return list_merger(weighting_dict, nRec)
def m_recommend(self, user_ids, nRec=10):
results = []
for uid in user_ids:
results.append(self.s_recommend(uid, nRec))
return results
def list_weighter(self, tupleDict, nRec, extra, weighting='parab'):
"""
:param tupleDict : dict{(list_of_items, weight)}
assumes list_of_items is ordered from best rec
to worst rec
:param nRec : number of items to recommend
:param extra : number of extra_items to consider
in the lists
:param weighting : - "linear" 1st place 10, 2nd place 9 ...
10th place 1
- "parab" 1st place 10,.. 5th place 3.5 ...
10th place 1
:return list of nRec items weighted according to dict
"""
# initialize a dict with items as keys and starting value zero
result = {}
count_dict = {}
for tuple in tupleDict.values():
items = tuple[0]
for i in range(nRec + extra):
result[str(items[i])] = 0
count_dict[str(items[i])] = 0
# assign a score based on position
for tuple in tupleDict.values():
items = tuple[0]
weight = tuple[1]
# weighting logic
if weighting == 'linear':
for i in range(nRec + extra):
result[str(items[i])] += (nRec + extra - i) * weight
elif weighting == 'parab':
for i in range(nRec + extra):
result[str(
items[i])] += (0.1 * i**2 - 1.92 * i + nRec) * weight
elif weighting == 'avg':
for i in range(nRec + extra):
result[str(items[i])] += (nRec - i) / 3
elif weighting == 'count_par':
for i in range(nRec + extra):
count_dict[str(items[i])] += 1
for i in range(nRec + extra):
result[str(items[i])] += (0.1 * i ** 2 - 1.92 * i + nRec) * weight \
+ 4 * count_dict.get(str(items[i]))
else:
raise ValueError('Not a valid weighting logic')
# sort the dict
sorted_results = sorted(result.items(), key=itemgetter(1))
rec_items = [x[0] for x in sorted_results]
# flip to order by decreasing order
rec_items = rec_items[::-1]
# return only the topN recommendations
return np.array(rec_items[0:nRec]).astype(int)
def setEnables(self, enable_dict):
self.enableSVD = enable_dict.get('enableSVD')
self.enableSLIM = enable_dict.get('enableSLIM')
self.enableUSER = enable_dict.get('enableUSER', False)
self.enableP3A = enable_dict.get('enableP3A', False)
def _filter_seen(self, user_id, ranking):
user_profile = self.urm.getCSR()[user_id]
seen = user_profile.indices
unseen_mask = np.in1d(ranking, seen, assume_unique=True, invert=True)
return ranking[unseen_mask]
def normalize_row(self, recommended_items, method):
if method == 'max':
norm_factor = recommended_items.max()
if norm_factor == 0: norm_factor = 1
return recommended_items / norm_factor
elif method == 'sum':
norm_factor = recommended_items.sum()
if norm_factor == 0: norm_factor = 1
return recommended_items / norm_factor
elif method == "none":
return recommended_items
else:
raise ValueError('Not a valid normalization method')
def getSlimRow(self, user):
return self.urm.getCSR().getrow(user) * self.slim_sim
def remove_duplicates(self, ordered_list):
"""
:param ordered_list
:return: the ordered_list still ordered removed of duplicates
"""
seen = set()
seen_add = seen.add
return [x for x in ordered_list if not (x in seen or seen_add(x))]
class XGBoostRecommender():
def __init__(self, urm, urm_t, icm, icm2, enable_dict, urm_test):
self.urm = urm
self.n_users, self.n_items = urm.getCSR().shape
self.setEnables(enable_dict )
self.item_item = IIHybridRecommender(urm, icm, icm2)
self.item_item.fit(item_weight=0.4, cbf1_weight=0.25, cbf2_weight=0.1)
self.user = CollaborativeFiltering()
self.user.fit(urm_t, k=100, h=0, mode='user')
if self.enableSVD:
self.svd = SVDRecommender(urm, nf=385)
if self.enableP3A:
self.p3a = P3alpha(urm.getCSR())
self.p3a.fit(topK=80, alpha=1, min_rating=0, implicit=True, normalize_similarity=True)
if self.enableSLIM:
choice = 2
logFile = open("SLIM_BPR_Cython.txt", "a")
self.slim = SLIM_BPR_Cython(urm.getCSR(), recompile_cython=False, positive_threshold=0,
URM_validation=urm_test.getCSR(), final_model_sparse_weights=True,
train_with_sparse_weights=False)
self.slim.fit(epochs=100, validation_every_n=1, logFile=logFile, batch_size=5, topK=200,
sgd_mode="adagrad", learning_rate=0.075)
if self.enableLFM:
# LightFM
print("starting USER CF")
self.lfm = LightFMRecommender()
self.lfm.fit(urm, epochs=100)
print("USER CF finished")
def buildXGBoostMatrix(self, recommenders, n):
print("building XGBoost Matrix")
user_id_col = []
slim_rec_col = []
itit_rec_col = []
p3a_rec_col = []
svd_rec_col = []
user_rec_col = []
lfm_rec_col = []
prof_len_col = []
for user in range(self.n_users):
# Item Item
itit_rec = self.item_item.s_recommend(user, n).tolist()
user_id_col.extend(itit_rec)
itit_rec_col.extend([user] * len(itit_rec))
# User
user_rec = self.user.g(user, n)
user_id_col.extend(user_rec)
user_rec_col.extend([user] * len(user_rec))
# P3A
if self.enableP3A:
p3a_rec = self.p3a.s_recommend(user, n)
user_id_col.extend(p3a_rec)
p3a_rec_col.extend([user] * len(p3a_rec))
# SVD
if self.enableSVD:
svd_rec = self.svd.s_recommend(user, n)
user_id_col.extend(svd_rec)
svd_rec_col.extend([user] * len(svd_rec))
# LFM
if self.enableLFM:
lfm_rec = self.lfm.s_recommend(user, n)
user_id_col.extend(lfm_rec)
lfm_rec_col.extend([user] * len(lfm_rec))
# SLIM
if self.enableSLIM:
slim_rec = self.slim.s_recommend(user, n)
user_id_col.extend(slim_rec)
slim_rec_col.extend([user] * len(slim_rec))
# Profile Len
profileLen = len(self.urm.extractTracksFromPlaylist(user))
prof_len_col.extend([profileLen] * len(user_rec))
dict = {"user_id": user_id_col,
"itit_rec_id": itit_rec_col,
"user_rec_id": user_rec_col}
# "slim_rec_id": slim_rec_col,
# "p3a_rec_id": p3a_rec_col,
# "lfm_rec_id": lfm_rec_col,
# "svd_rec_id": svd_rec_col,
# "profile_len": prof_len_col}
self.buildDataFrame(dict)
def setEnables(self, enable_dict):
self.enableSVD = enable_dict.get('enableSVD')
self.enableSLIM = enable_dict.get('enableSLIM')
self.enableCBF2 = enable_dict.get('enableCBF2')
self.enableP3A = enable_dict.get('enableP3A')
self.enableLFM = enable_dict.get('enableLFM')
def buildDataFrame(self, dict):
print("building dataframe")
self.df = pd.DataFrame(dict)
self.df.describe()
print("built df")