def test_tf_bmf_batch_accuracy(tf_session):
from lenskit.algorithms import basic
from lenskit.algorithms import bias
import lenskit.crossfold as xf
from lenskit import batch
import lenskit.metrics.predict as pm
ratings = lktu.ml100k.ratings
algo = lktf.BiasedMF(25,
damping=10,
batch_size=1024,
epochs=20,
rng_spec=42)
algo = basic.Fallback(algo, bias.Bias(damping=10))
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
return batch.predict(algo, test)
folds = xf.partition_users(ratings, 5, xf.SampleFrac(0.2))
preds = pd.concat(eval(train, test) for (train, test) in folds)
mae = pm.mae(preds.prediction, preds.rating)
assert mae == approx(0.83, abs=0.025)
user_rmse = preds.groupby('user').apply(
lambda df: pm.rmse(df.prediction, df.rating))
assert user_rmse.mean() == approx(1.03, abs=0.05)
def test_fsvd_batch_accuracy():
from lenskit.algorithms import basic
from lenskit.algorithms import bias
import lenskit.crossfold as xf
from lenskit import batch
import lenskit.metrics.predict as pm
ratings = lktu.ml100k.ratings
svd_algo = svd.FunkSVD(25, 125, damping=10)
algo = basic.Fallback(svd_algo, bias.Bias(damping=10))
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
return batch.predict(algo, test)
folds = xf.partition_users(ratings, 5, xf.SampleFrac(0.2))
preds = pd.concat(eval(train, test) for (train, test) in folds)
mae = pm.mae(preds.prediction, preds.rating)
assert mae == approx(0.74, abs=0.025)
user_rmse = preds.groupby('user').apply(lambda df: pm.rmse(df.prediction, df.rating))
assert user_rmse.mean() == approx(0.92, abs=0.05)
def test_batch_rmse():
import lenskit.crossfold as xf
import lenskit.batch as batch
import lenskit.algorithms.bias as bs
ratings = lktu.ml100k.ratings
algo = bs.Bias(damping=5)
def eval(train, test):
algo.fit(train)
preds = batch.predict(algo, test)
return preds.set_index(['user', 'item'])
results = pd.concat(
(eval(train, test)
for (train, test) in xf.partition_users(ratings, 5, xf.SampleN(5))))
user_rmse = results.groupby('user').apply(
lambda df: pm.rmse(df.prediction, df.rating))
# we should have all users
users = ratings.user.unique()
assert len(user_rmse) == len(users)
missing = np.setdiff1d(users, user_rmse.index)
assert len(missing) == 0
# we should not have any missing values
assert all(user_rmse.notna())
# we should have a reasonable mean
assert user_rmse.mean() == approx(0.93, abs=0.05)
def test_bias_batch_predict(ncpus):
from lenskit.algorithms import bias
import lenskit.crossfold as xf
from lenskit import batch
import lenskit.metrics.predict as pm
ratings = lktu.ml100k.ratings
algo = bias.Bias(damping=5)
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
recs = batch.predict(algo, test, n_jobs=ncpus)
return recs
preds = pd.concat(
(eval(train, test)
for (train,
test) in xf.partition_users(ratings, 5, xf.SampleFrac(0.2))))
_log.info('analyzing predictions')
rmse = pm.rmse(preds.prediction, preds.rating)
_log.info('RMSE is %f', rmse)
assert rmse == pytest.approx(0.95, abs=0.1)
def test_batch_predict_preshared():
"Test batch prediction with isolated training and a pre-serialized algorithm."
from lenskit.algorithms import bias
import lenskit.crossfold as xf
algo = bias.Bias()
splits = xf.sample_users(lktu.ml_test.ratings, 1, 100, xf.SampleN(5))
train, test = next(splits)
ares = lkb.train_isolated(algo, train)
preds = lkb.predict(ares, test)
assert len(preds) == len(test)
assert not any(preds['prediction'].isna())
def test_uu_batch_accuracy():
from lenskit.algorithms import basic
from lenskit.algorithms import bias
import lenskit.crossfold as xf
import lenskit.metrics.predict as pm
ratings = lktu.ml100k.ratings
uu_algo = knn.UserUser(30)
algo = basic.Fallback(uu_algo, bias.Bias())
folds = xf.partition_users(ratings, 5, xf.SampleFrac(0.2))
preds = [__batch_eval((algo, train, test)) for (train, test) in folds]
preds = pd.concat(preds)
mae = pm.mae(preds.prediction, preds.rating)
assert mae == approx(0.71, abs=0.05)
user_rmse = preds.groupby('user').apply(lambda df: pm.rmse(df.prediction, df.rating))
assert user_rmse.mean() == approx(0.91, abs=0.055)
def test_topn_big():
ratings = lktu.ml_test.ratings
users = ratings.user.unique()
items = ratings.item.unique()
user_items = ratings.set_index('user').item
algo = basic.TopN(bias.Bias())
a2 = algo.fit(ratings)
assert a2 is algo
# test 100 random users
for u in np.random.choice(users, 100, False):
recs = algo.recommend(u, 100)
assert len(recs) == 100
rated = user_items.loc[u]
assert all(~recs['item'].isin(rated))
unrated = np.setdiff1d(items, rated)
scores = algo.predictor.predict_for_user(u, unrated)
top = scores.nlargest(100)
assert top.values == approx(recs.score.values)
