def get_algs():
# create a dict of (textual algorithm description => class) to be evaluated
algs = {}
ara = ar.AssosiationRules()
algs['ar'] = ara
del ara
sra = sr.SequentialRules(10, weighting='div', pruning=20)
algs['sr10-div'] = sra
del sra
sknna = sknn.ContextKNN(500, 1000, similarity="cosine")
algs['sknn-500-1000-cosine'] = sknna
del sknna
ssknna = ssknn.SeqContextKNN(100, 500, similarity="cosine")
algs['ssknn-100-500-cosine'] = ssknna
del ssknna
knn = sfsknn.SeqFilterContextKNN(100, 500, similarity="cosine")
algs['sfknn-100-500-cosine'] = knn
del knn
vsknna = vsknn.VMContextKNN(100, 1000, similarity="cosine")
algs['svmknn-100-1000-cosine'] = vsknna
vsknna = vsknn.VMContextKNN(100, 2000, similarity="cosine")
algs['svmknn-100-2000-cosine'] = vsknna
del vsknna
grunew = gru4rec2.GRU4Rec(loss='bpr-max-0.5',
final_act='linear',
hidden_act='tanh',
layers=[100],
batch_size=32,
dropout_p_hidden=0.0,
learning_rate=0.2,
momentum=0.5,
n_sample=2048,
sample_alpha=0,
time_sort=True)
algs['gru-100-bpr-max-0.5'] = grunew
del grunew
return algs
metric.append(ac.HitRate(5))
metric.append(ac.HitRate(10))
metric.append(ac.HitRate(20))
metric.append(ac.MRR(5))
metric.append(ac.MRR(10))
metric.append(ac.MRR(20))
metric.append(ac.NDCG(5))
metric.append(ac.NDCG(10))
metric.append(ac.NDCG(20))
# create a dict of (textual algorithm description => class) to be evaluated
algs = {}
#weighted hybrid example
hybrid = wh.WeightedHybrid([
cknn.ContextKNN(100, 500, similarity="cosine"),
gru4rec.GRU4Rec(layers=[100], n_sample=128, sample_alpha=0)
], [0.3, 0.7]) # weights
algs['whybrid-test-10-90'] = hybrid
train, test = loader.load_data(data_path,
file_prefix,
rows_train=limit_train,
rows_test=limit_test,
density=density_value)
item_ids = train.ItemId.unique()
# init metrics
for m in metric:
m.init(train)
sra = sr.SequentialRules(10,
weighting='div',
pruning=20,
last_n_days=None)
algs['sr10-div'] = sra
ara = ar.AssosiationRules()
algs['ar'] = ara
#knn
iknna = iknn.ItemKNN()
algs['iknn'] = iknna
sknna = sknn.ContextKNN(100, 500, similarity="cosine", extend=False)
algs['sknn-100-500-cosine'] = sknna
vsknna = vsknn.VMContextKNN(100,
2000,
similarity="cosine",
last_n_days=None,
extend=False)
algs['vsknn-200-2000-cosine'] = vsknna
ssknna = ssknn.SeqContextKNN(100,
500,
similarity="cosine",
extend=False)
algs['ssknn-100-500-cosine-div'] = ssknna
algs['gru-100-top1'] = gru
if method == 'gru4rec2':
gru = gru4rec.GRU4Rec( loss='bpr-max-0.5', final_act='tanh', hidden_act='tanh', layers=[1000], dropout_p_hidden=0.1, learning_rate=0.05, momentum=0.1, n_sample=2048, sample_alpha=0)
algs['gru-100-bpr-max-0.5'] = gru
if method == 'iknn':
# item knn example
knn = iknn.ItemKNN()
algs['iknn'] = knn
if method == 'ct':
# item knn example
ctt = ct.ContextTree()
algs['CT'] = ctt
isCT = True
if method == 'cknn':
# context knn example
knn = cknn.ContextKNN()
algs['knn-500-1000-cosine'] = knn
if method == 'scknn':
scknn_obj = scknn.SeqContextKNN(100, 500, similarity="cosine", extend=False)
algs['scknn-100-500-cosine-div'] = scknn_obj
if method == 'smf':
smf_obj = smf.SessionMF(factors=100, batch=50, learn='adagrad_sub', learning_rate=0.085, momentum=0.2,
regularization=0.005, dropout=0.3, skip=0.0, epochs=10, shuffle=-1,
activation='linear', objective='top1_max', samples=2048)
algs['smf-bpr'] = smf_obj
'''
Execution
'''
#load data
train, test = loader.load_data( data_path, file_prefix, rows_train=limit_train, rows_test=limit_test, density=density_value)
item_ids = train.ItemId.unique()