def test_implicit_bpr_train_rec():
algo = BPR(25)
ratings = lktu.ml_pandas.renamed.ratings
algo.fit(ratings)
recs = algo.recommend(100, n=20)
assert len(recs) == 20
def test_implicit_bpr_train_rec():
algo = BPR(25)
assert algo.factors == 25
ratings = lktu.ml_pandas.renamed.ratings
algo.fit(ratings)
recs = algo.recommend(100, n=20)
assert len(recs) == 20
_log.info('serializing implicit model')
mod = pickle.dumps(algo)
_log.info('serialized to %d bytes')
a2 = pickle.loads(mod)
r2 = a2.recommend(100, n=20)
assert len(r2) == 20
assert all(r2 == recs)
def test_implicit_pickle_untrained(tmp_path):
mf = tmp_path / 'bpr.dat'
algo = BPR(25)
with mf.open('wb') as f:
pickle.dump(algo, f)
with mf.open('rb') as f:
a2 = pickle.load(f)
assert a2 is not algo
assert a2.factors == 25
def get_topn_algo_class(algo):
if algo == 'popular':
return basic.Popular()
elif algo == 'bias':
return basic.TopN(basic.Bias())
elif algo == 'itemitem':
return basic.TopN(iknn.ItemItem(nnbrs=-1, center=False, aggregate='sum'))
elif algo == 'useruser':
return basic.TopN(uknn.UserUser(nnbrs=5, center=False, aggregate='sum'))
elif algo == 'biasedmf':
return basic.TopN(als.BiasedMF(50, iterations=10))
elif algo == 'implicitmf':
return basic.TopN(als.ImplicitMF(20, iterations=10))
elif algo == 'funksvd':
return basic.TopN(svd.FunkSVD(20, iterations=20))
elif algo == 'bpr':
return basic.TopN(BPR(25))
def test_implicit_bpr_train_rec():
algo = BPR(25, use_gpu=False)
assert algo.factors == 25
ratings = lktu.ml_test.ratings
algo.fit(ratings)
recs = algo.recommend(100, n=20)
assert len(recs) == 20
preds = algo.predict_for_user(100, [20, 30, 23148010])
assert all(preds.index == [20, 30, 23148010])
assert all(preds.isna() == [False, False, True])
_log.info('serializing implicit model')
mod = pickle.dumps(algo)
_log.info('serialized to %d bytes')
a2 = pickle.loads(mod)
r2 = a2.recommend(100, n=20)
assert len(r2) == 20
assert all(r2 == recs)
pruned_data_5.reset_index(inplace=True)
#pairs_user = list(partition_users(pruned_data_5, 5, xf.SampleN(1)))
pairs_user = list(sample_users(pruned_data_5, 5, 12000, xf.SampleN(1)))
pickle_out = open("sample_user.pickle", "wb")
pickle.dump(pairs_user, pickle_out)
pickle_out.close()
truth = pd.concat((p.test for p in pairs_user))
#truth.to_csv(r'results/steam/pruned_5.csv')
def algo_eval(path, algo, dataset):
evaluation = batch.MultiEval(path=path, predict=False, recommend=100)
evaluation.add_algorithms(algos=algo)
evaluation.add_datasets(data=dataset)
evaluation.run()
algo_ii = item_knn.ItemItem(20, center=False, aggregate='sum')
#algo_uu = user_knn.UserUser(30, center=False, aggregate='sum')
algo_pop = basic.Popular()
algo_mf = ImplicitMF(40)
algo_bpr = BPR()
algo_tf_idf = tf_idf()
algo_LDA = LDA()
algo_eval('results/steam/all_algo_sample_user',
[algo_LDA, algo_tf_idf, algo_ii, algo_pop, algo_mf, algo_bpr],
pairs_user)
item_to_item_100 = item_knn.ItemItem(100)
item_to_item_200 = item_knn.ItemItem(200)
item_to_item_500 = item_knn.ItemItem(500)
user_to_user_100 = user_knn.UserUser(100)
user_to_user_200 = user_knn.UserUser(200)
user_to_user_500 = user_knn.UserUser(500)
biased_mf_50 = als.BiasedMF(50)
biased_mf_100 = als.BiasedMF(100)
biased_mf_200 = als.BiasedMF(200)
implicit_mf_50 = als.ImplicitMF(50)
implicit_mf_100 = als.ImplicitMF(100)
implicit_mf_200 = als.ImplicitMF(200)
funk_svd_mf_50 = funksvd.FunkSVD(50)
funk_svd_mf_100 = funksvd.FunkSVD(100)
funk_svd_mf_200 = funksvd.FunkSVD(200)
bayesian = BPR()
hierarchical_poisson_fact_50 = HPF(50)
hierarchical_poisson_fact_100 = HPF(100)
hierarchical_poisson_fact_200 = HPF(200)
train, test = train_test_split(ratings[['user', 'item', 'rating']],
test_size=0.2)
eval = batch.MultiEval('../recs/cf', recommend=NUM_OF_RECS)
eval.add_datasets((train, test), name='ml-1m')
eval.add_algorithms(random, name='random')
eval.add_algorithms(popular, name='popular')
eval.add_algorithms(item_to_item_100, name='item_to_item_100')
eval.add_algorithms(item_to_item_200, name='item_to_item_200')
eval.add_algorithms(item_to_item_500, name='item_to_item_500')
eval.add_algorithms(user_to_user_100, name='user_to_user_100')
def default(implicit):
return BPR(50)
def instantiate(opts, implicit):
feats, reg, lrate = opts
return BPR(feats, regularization=reg, learning_rate=lrate)
"""
Code for optimizing the item-item algorithm.
"""
from skopt.space import Integer, Real
from lenskit.algorithms.implicit import BPR
from bookgender.config import have_gpu
dimensions = [Integer(5, 250), Real(0, 0.25), Real(0, 0.1)]
def instantiate(opts, implicit):
feats, reg, lrate = opts
return BPR(feats, regularization=reg, learning_rate=lrate)
def default(implicit):
return BPR(50)
update = None
sweep_points = [
BPR(factors=nf, use_gpu=have_gpu()) for nf in range(25, 250, 25)
]
sweep_attrs = ['factors']