def fit(self,
topK=600,
shrink=1000,
similarity='asymmetric',
normalize=True,
feature_weighting="BM25",
save_model=False,
best_parameters=False,
**similarity_args):
similarity_args = {'asymmetric_alpha': 0.40273209903969387}
if best_parameters:
m = OfflineDataLoader()
folder_path_icbf, file_name_icbf = m.get_parameter(
self.RECOMMENDER_NAME)
self.loadModel(folder_path=folder_path_icbf,
file_name=file_name_icbf)
if self.feature_weighting == "none":
pass
if self.feature_weighting == "BM25":
self.ICM = self.ICM.astype(np.float32)
self.ICM = to_okapi(self.ICM)
elif self.feature_weighting == "TF-IDF":
self.ICM = self.ICM.astype(np.float32)
self.ICM = to_tfidf(self.ICM)
similarity = Compute_Similarity(self.ICM.T,
shrink=shrink,
topK=topK,
normalize=normalize,
similarity=similarity,
**similarity_args)
else:
self.topK = topK
self.shrink = shrink
similarity = Compute_Similarity(self.ICM.T,
shrink=shrink,
topK=topK,
normalize=normalize,
similarity=similarity,
**similarity_args)
self.parameters = "sparse_weights= {0}, similarity= {1}, shrink= {2}, neighbourhood={3}, " \
"normalize= {4}".format(
self.sparse_weights, similarity, self.shrink, self.topK, normalize)
if self.sparse_weights:
self.W_sparse = similarity.compute_similarity()
else:
self.W = similarity.compute_similarity()
self.W = self.W.toarray()
if save_model:
self.saveModel("saved_models/submission/",
file_name="ItemKNNCBFRecommender_submission_model")
def fit(self, topK=250, shrink=100, alpha=0.7017094, beta=0.51034483, gamma=0.16206897, normalize=False,
similarity="jaccard", **similarity_args):
self.k = topK
self.shrink = shrink
self.alpha = alpha
self.beta = beta
self.gamma = gamma
self.normalize = normalize
print("Item Tree Hybrid Recommender: Model fitting begins")
# Calculate all the Similarity Matrices One by one
# URM tfidf --> 50446 x 50446
self.sim_URM_tfidf = Compute_Similarity(self.URM_train_tfidf.T, shrink=0, topK=200, normalize=normalize,
similarity=similarity, **similarity_args)
# ICM tfidf --> 20635 x 20635
self.ICM = to_okapi(self.ICM)
self.sim_ICM_tfidf = Compute_Similarity(self.ICM.T, shrink=0, topK=25, normalize=normalize,
similarity=similarity, **similarity_args)
# URM.T tfidf --> 20635 x 20635
self.sim_URM_T_tfidf = Compute_Similarity(self.URM_train_tfidf, shrink=10, topK=350, normalize=normalize,
similarity=similarity, **similarity_args)
# Slim --> 20635 x 20635
self.sim_Slim = Slim_mark1(self.URM_train)
if self.sparse_weights:
# URM
self.W_sparse_URM = self.sim_URM_tfidf.compute_similarity()
# UCM
self.W_sparse_ICM = self.sim_ICM_tfidf.compute_similarity()
# self.W_sparse_UCM = self.sim_UCM_tfidf.fit()
# ICM
self.W_sparse_URM_T = self.sim_URM_T_tfidf.compute_similarity()
# Slim
# lambda_i = 0.37142857, lambda_j = 0.97857143
self.W_sparse_Slim = self.sim_Slim.fit()
# add the parameters for the logging
self.parameters = "sparse_weights= {}, similarity= {},shrink= {}, neighbourhood={},normalize= {}, alpha= {}, beta={}, gamma={}".format(
self.sparse_weights, similarity, shrink, topK, normalize,
alpha, beta, gamma)
def fit(self,
topK=175,
shrink=400,
similarity="asymmetric",
normalize=True,
feature_weighting="BM25",
save_model=False,
best_parameters=False,
location="training",
submission=False,
offline=False,
**similarity_args):
if offline:
m = OfflineDataLoader()
folder_path_icf, file_name_icf = m.get_model(
self.RECOMMENDER_NAME, training=(not submission))
self.loadModel(folder_path=folder_path_icf,
file_name=file_name_icf)
else:
if best_parameters:
m = OfflineDataLoader()
folder_path_ucf, file_name_ucf = m.get_parameter(
self.RECOMMENDER_NAME)
self.loadModel(folder_path=folder_path_ucf,
file_name=file_name_ucf)
if self.feature_weighting == "none":
similarity = Compute_Similarity(self.URM_train.T,
**similarity_args)
else:
if feature_weighting == "BM25":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_okapi(self.URM_train)
elif feature_weighting == "TF-IDF":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_tfidf(self.URM_train)
similarity_args = {
'asymmetric_alpha': 0.11483114799990246,
'normalize': True,
'shrink': 450,
'similarity': 'asymmetric',
'topK': 200
}
similarity = Compute_Similarity(self.URM_train_copy.T,
**similarity_args)
else:
self.topK = topK
self.shrink = shrink
self.feature_weighting = feature_weighting
similarity_args = {'asymmetric_alpha': 0.0033404951135529437}
if self.feature_weighting == "BM25":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_okapi(self.URM_train)
elif self.feature_weighting == "TF-IDF":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_tfidf(self.URM_train)
if self.feature_weighting == "none":
similarity = Compute_Similarity(self.URM_train.T,
shrink=shrink,
topK=topK,
normalize=normalize,
similarity=similarity,
**similarity_args)
else:
similarity = Compute_Similarity(self.URM_train_copy.T,
shrink=shrink,
topK=topK,
normalize=normalize,
similarity=similarity,
**similarity_args)
self.parameters = "sparse_weights= {0}, similarity= {1}, shrink= {2}, neighbourhood={3}, " \
"normalize= {4}".format(self.sparse_weights, similarity, shrink, topK, normalize)
if self.sparse_weights:
self.W_sparse = similarity.compute_similarity()
else:
self.W = similarity.compute_similarity()
self.W = self.W.toarray()
if save_model:
self.saveModel("saved_models/submission/",
file_name=self.RECOMMENDER_NAME + "_" + location +
"_model")
def get_URM_train_okapi(self):
if self.URM_train is None:
raise TypeError("PlaylistDataReader: URM train is not build")
else:
self.URM_train_okapi = to_okapi(self.URM_train)
return self.URM_train_okapi
def fit(self,
topK=400,
shrink=200,
similarity='cosine',
feature_weighting="BM25",
normalize=True,
save_model=False,
best_parameters=False,
offline=False,
submission=False,
location="submission",
**similarity_args):
#similarity_args = {'tversky_alpha': 0.8047100184165605, 'tversky_beta': 1.9775806370926445}
#self.feature_weighting = feature_weighting
if offline:
m = OfflineDataLoader()
folder_path_icf, file_name_icf = m.get_model(
self.RECOMMENDER_NAME, training=(not submission))
self.loadModel(folder_path=folder_path_icf,
file_name=file_name_icf)
else:
if best_parameters:
m = OfflineDataLoader()
folder_path_icf, file_name_icf = m.get_parameter(
self.RECOMMENDER_NAME)
self.loadModel(folder_path=folder_path_icf,
file_name=file_name_icf)
#similarity_args = {'normalize': True, 'shrink': 0, 'similarity': 'tversky', 'topK': 20, 'tversky_alpha': 0.18872151621891953, 'tversky_beta': 1.99102432161935}
similarity_args = {
'feature_weighting': 'BM25',
'normalize': True,
'shrink': 200,
'similarity': 'cosine',
'topK': 400
}
if self.feature_weighting == "none":
pass
if self.feature_weighting == "BM25":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_okapi(self.URM_train)
elif self.feature_weighting == "TF-IDF":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_tfidf(self.URM_train)
similarity = Compute_Similarity(self.URM_train_copy,
**similarity_args)
else:
self.topK = topK
self.shrink = shrink
self.feature_weighting = feature_weighting
if self.feature_weighting == "BM25":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_okapi(self.URM_train)
elif self.feature_weighting == "TF-IDF":
self.URM_train_copy = self.URM_train.astype(np.float32)
self.URM_train_copy = to_tfidf(self.URM_train)
if self.feature_weighting == "none":
similarity = Compute_Similarity(self.URM_train,
shrink=shrink,
topK=topK,
normalize=normalize,
similarity=similarity,
**similarity_args)
else:
similarity = Compute_Similarity(self.URM_train_copy,
shrink=shrink,
topK=topK,
normalize=normalize,
similarity=similarity,
**similarity_args)
self.parameters = "sparse_weights= {0}, similarity= {1}, shrink= {2}, neighbourhood={3}, normalize={4}".format(
self.sparse_weights, similarity, shrink, topK, normalize)
if self.sparse_weights:
self.W_sparse = similarity.compute_similarity()
else:
self.W = similarity.compute_similarity()
self.W = self.W.toarray()
if save_model:
self.saveModel("saved_models/" + location + "/",
file_name=self.RECOMMENDER_NAME + "_" + location +
"_model")