def test_tf_bpr_general(tmp_path, tf_session):
"Training, saving, loading, and using a BPR model."
fn = tmp_path / 'bias.bpk'
ratings = lktu.ml_test.ratings
original = lktf.BPR(20,
batch_size=1024,
epochs=20,
neg_count=2,
rng_spec=42)
original.fit(ratings)
ue = original.model.get_layer('user-embed')
assert ue.get_weights()[0].shape == (ratings.user.nunique(), 20)
ie = original.model.get_layer('item-embed')
assert ie.get_weights()[0].shape == (ratings.item.nunique(), 20)
binpickle.dump(original, fn)
_log.info('serialized to %d bytes', fn.stat().st_size)
algo = binpickle.load(fn)
# does scoring work?
preds = algo.predict_for_user(100, [5, 10, 30])
assert all(preds.notna())
# can we include a nonexistent item?
preds = algo.predict_for_user(100, [5, 10, 230413804])
assert len(preds) == 3
assert all(preds.loc[[230413804]].isna())
assert preds.isna().sum() == 1
def test_tf_bpr_batch_accuracy(tf_session):
from lenskit.algorithms import basic
import lenskit.crossfold as xf
from lenskit import batch, topn
ratings = lktu.ml100k.ratings
algo = lktf.BPR(20, batch_size=1024, epochs=20, rng_spec=42)
algo = Recommender.adapt(algo)
all_recs = []
all_test = []
for train, test in xf.partition_users(ratings, 5, xf.SampleFrac(0.2)):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
recs = batch.recommend(algo, np.unique(test.user), 50)
all_recs.append(recs)
all_test.append(test)
_log.info('analyzing results')
rla = topn.RecListAnalysis()
rla.add_metric(topn.ndcg)
rla.add_metric(topn.recip_rank)
scores = rla.compute(pd.concat(all_recs, ignore_index=True),
pd.concat(all_test, ignore_index=True),
include_missing=True)
scores.fillna(0, inplace=True)
_log.info('MRR: %f', scores['recip_rank'].mean())
_log.info('nDCG: %f', scores['ndcg'].mean())
assert scores['ndcg'].mean() > 0.1
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))
elif algo == 'tf_bpr':
return basic.TopN(
lktf.BPR(20, batch_size=1024, epochs=5, neg_count=2, rng_spec=42))
def get_algo_class(algo):
if algo == 'popular':
return basic.Popular()
elif algo == 'bias':
return basic.Bias(users=False)
elif algo == 'topn':
return basic.TopN(basic.Bias())
elif algo == 'itemitem':
return iknn.ItemItem(nnbrs=-1)
elif algo == 'useruser':
return uknn.UserUser(nnbrs=5)
elif algo == 'biasedmf':
return als.BiasedMF(50, iterations=10)
elif algo == 'implicitmf':
return als.ImplicitMF(20, iterations=10)
elif algo == 'funksvd':
return svd.FunkSVD(20, iterations=20)
elif algo == 'tf_bpr':
return lktf.BPR(20,
batch_size=1024,
epochs=5,
neg_count=2,
rng_spec=42)
movie_titles["vector"] = np.nan
movie_titles["vector"] = movie_titles["vector"].astype('object')
for index, item in enumerate(movie_titles["title"]):
vector = np.mean(np.array([new_model[x] for x in item]), axis=0)
vector = np.concatenate([vector, one_hot[index]])
movie_titles.at[index, 'vector'] = np.ndarray.tolist(vector)
movie_titles["title"] = movie_titles["title"].apply(lambda x: " ".join(x))
movie_titles = pd.merge(movie_titles, output_df, on="title")
movie_titles = movie_titles.drop(columns=["genres"])
movie_titles.rename(columns={"vector": "item", "userId": "user"}, inplace=True)
movie_titles["item"] = movie_titles["item"].apply(lambda x: str(x))
print(movie_titles.head())
net = tf.BPR(features=30, epochs=80)
net = net.fit(movie_titles.reset_index(drop=True))
un_user = output_df.userId.unique()
result = pd.Series()
full_result = pd.DataFrame()
i = 0
printProgressBar(0,
len(un_user),
prefix='Progress:',
suffix='Complete',
length=50)
for user in un_user:
result = pd.DataFrame(
""" This module defines the algorithms, and their default configurations, that we are going to use. """ from lenskit.algorithms import item_knn, user_knn, als, tf from lenskit.algorithms import basic Bias = basic.Bias(damping=5) Pop = basic.Popular() II = item_knn.ItemItem(20, save_nbrs=2500) UU = user_knn.UserUser(30) ALS = als.BiasedMF(50) IALS = als.ImplicitMF(50) BPR = tf.BPR(50) TFMF = tf.IntegratedBiasMF(50)