def get_recommendations_KNN(user_id, nb_recommendations = 1):
'''
Return a recommendation
:param user: id of user you want a recommendation
:param algorithm: "Funk"
:return:
'''
algo = algoKNN
model = modelKNN
recs = batch.recommend(algo, users,nb_recommendations, topn.UnratedCandidates(train), nprocs=None)
recs = recs[recs['user'] == user_id]
if not recs.empty:
# select colums
rows = recs[['item', 'score']]
rows_dict = rows.to_dict(orient="records")
recommendations = []
for entry in rows_dict:
recommendations.append({
"item": int(entry['item']),
"score": int(entry["score"])
})
# otherwise, nope, not found
else:
abort(
404, "No ratings for user with user_id {user_id} ".format(user_id=user_id)
)
return recommendations
def test_uu_implicit_batch_accuracy():
from lenskit import batch, topn
import lenskit.crossfold as xf
ratings = lktu.ml100k.ratings
algo = knn.UserUser(30, center=False, aggregate='sum')
folds = list(xf.partition_users(ratings, 5, xf.SampleFrac(0.2)))
all_test = pd.concat(f.test for f in folds)
rec_lists = []
for train, test in folds:
_log.info('running training')
algo.fit(train.loc[:, ['user', 'item']])
cands = topn.UnratedCandidates(train)
_log.info('testing %d users', test.user.nunique())
recs = batch.recommend(algo, test.user.unique(), 100, cands, n_jobs=2)
rec_lists.append(recs)
recs = pd.concat(rec_lists)
rla = topn.RecListAnalysis()
rla.add_metric(topn.ndcg)
results = rla.compute(recs, all_test)
user_dcg = results.ndcg
dcg = user_dcg.mean()
assert dcg >= 0.03
def get_topn(user,num_recommendations=10): #added argument num_recommendations=10
#algo = algoKNN
#model = modelKNN
# merged the code for get_topn and getRecommendaions into one function:
recs = batch.recommend(algo, users,num_recommendations, topn.UnratedCandidates(train), nprocs=None) # changed arguments for batch.recommend
recs = recs[recs['user'] == user]
if not recs.empty:
# select colums
rows = recs[['item', 'score']]
rows_dict = rows.to_dict(orient="records")
recommendations = []
for entry in rows_dict:
recommendations.append({
"item": int(entry['item']),
"score": int(entry["score"])
})
# otherwise, nope, not found
else:
abort(
404, "No ratings for user with user_id {user_id} ".format(user_id=user) #user_id=user_id was changed to user_id=user
)
return recommendations
def eval(train, test):
_log.info('running training')
train['rating'] = train.rating.astype(np.float_)
algo.fit(train)
users = test.user.unique()
_log.info('testing %d users', len(users))
candidates = topn.UnratedCandidates(train)
recs = batch.recommend(algo, users, 100, candidates)
return recs
def test_unrated():
ratings = lktu.ml_pandas.renamed.ratings
unrate = topn.UnratedCandidates(ratings)
cs = unrate(100)
items = ratings.item.unique()
rated = ratings[ratings.user == 100].item.unique()
assert len(cs) == len(items) - len(rated)
assert len(np.intersect1d(cs, rated)) == 0
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
cand_fun = topn.UnratedCandidates(train)
recs = batch.recommend(algo,
test.user.unique(),
100,
cand_fun,
n_jobs=ncpus)
return recs
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
cand_fun = topn.UnratedCandidates(train)
recs = batch.recommend(algo, test.user.unique(), 100, cand_fun)
# combine with test ratings for relevance data
res = pd.merge(recs, test, how='left', on=('user', 'item'))
# fill in missing 0s
res.loc[res.rating.isna(), 'rating'] = 0
return res
def eval(train, test):
train['rating'] = train.rating.astype(np.float_)
_log.info('training CG')
cg_algo.fit(train)
_log.info('training LU')
lu_algo.fit(train)
users = test.user.unique()
_log.info('testing %d users', len(users))
candidates = topn.UnratedCandidates(train)
cg_recs = batch.recommend(cg_algo, users, 100, candidates, n_jobs=2)
lu_recs = batch.recommend(lu_algo, users, 100, candidates, n_jobs=2)
return pd.concat({
'CG': cg_recs,
'LU': lu_recs
}, names=['Method']).reset_index('Method')
def test_uu_implicit_batch_accuracy():
from lenskit import batch, topn
import lenskit.crossfold as xf
import lenskit.metrics.topn as lm
ratings = lktu.ml100k.load_ratings()
algo = knn.UserUser(30, center=False, aggregate='sum')
folds = xf.partition_users(ratings, 5, xf.SampleFrac(0.2))
rec_lists = []
for train, test in folds:
_log.info('running training')
algo.fit(train.loc[:, ['user', 'item']])
cands = topn.UnratedCandidates(train)
_log.info('testing %d users', test.user.nunique())
recs = batch.recommend(algo, test.user.unique(), 100, cands, test)
rec_lists.append(recs)
recs = pd.concat(rec_lists)
user_dcg = recs.groupby('user').rating.apply(lm.dcg)
dcg = user_dcg.mean()
assert dcg >= 0.1
recs['Algorithm'] = aname
return recs """
all_recs = []
test_data = []
for train, test in xf.partition_users(ratings[['user', 'item', 'rating']], 5,
xf.SampleFrac(0.2)):
test_data.append(test)
# ItemItem
# eval('ItemItem', algo_ii, train, test)
#fittable = util.clone(algo)
algo_ii.fit(train)
users = test.user.unique()
# the recommend function can merge rating values
recs = batch.recommend(algo_ii, users, 100, topn.UnratedCandidates(train),
test)
# add the algorithm
recs['Algorithm'] = 'ItemItem'
all_recs.append(recs)
# ItemItem
# all_recs.append(eval('ALS', algo_als, train, test))
#fittable = util.clone(algo)
algo_als.fit(train)
users2 = test.user.unique()
# the recommend function can merge rating values
recs2 = batch.recommend(algo_als, users2, 100,
topn.UnratedCandidates(train), test)
# add the algorithm
def eval(train, test):
model = algo.fit(train)
users = test.user.unique()
recs = batch.recommend(model, users, 20, topn.UnratedCandidates(train))
print('Recommended to test set of users.')
return recs
print(modelAls.user_index_[0:10])
first = modelAls.user_index_[0]
print(type(first))
firstUser = np.array([first], np.int64)
# Get recommendation for a user
# As als.BiasedMF is not implementing Recommend
# We need to first adapt the model
# https://lkpy.lenskit.org/en/stable/interfaces.html#recommendation
# rec = Recommender.adapt(modelAls)
# recs = rec.recommend(first, 10) #Gives error
# Get 10 recommendations for a user (pandas dataframe)
recs = batch.recommend(modelAls, firstUser,
10, topn.UnratedCandidates(train), test)
print(recs)
# Get the first recommended item
firstRec = recs.iloc[0, 0]
firstRecScore = recs.iloc[0, 1]
print(firstRec)
# Get the explanation of the recommendation
# Get the index of the items
items = modelAls.item_index_
# Find the index of the first item
indexFirstRec = items.get_loc(firstRec)
# Get the feature values of the user
userProfile = modelAls.user_features_[0]
itemProfile = modelAls.item_features_[indexFirstRec]
