from reco.recommender import FunkSVD
from reco.metrics import rmse
from reco.datasets import loadMovieLens100k

train, test, _, _ = loadMovieLens100k(train_test_split=True)

f = FunkSVD(k=64, learning_rate=0.002, regularizer = 0.05, iterations = 150, method = 'stochastic', bias=True)

# fits the model to the data
f.fit(X=train, formatizer={'user':'userId', 'item':'itemId', 'value':'rating'},verbose=True)

# predicts the ratings from the test set
preds = f.predict(X=test, formatizer={'user':'userId', 'item':'itemId'})
print(rmse(preds, list(test['rating']))

from reco.recommender import SVDRecommender
from reco.datasets import loadMovieLens100k
from reco.metrics import rmse

train, test, _, _ = loadMovieLens100k(train_test_split=True)

svd = SVDRecommender(no_of_features=8)

# Creates the user-item matrix, the userIds on the rows and the itemIds on the columns.
user_item_matrix, users, items = svd.create_utility_matrix(train, formatizer={'user':'userId', 'item':'itemId', 'value':'rating'})

# fits the svd model to the matrix data.
svd.fit(user_item_matrix, users, items)

# predict the ratings from test set
preds = svd.predict(test, formatizer = {'user':'userId', 'item': 'itemId'})
print(rmse(preds, list(test['rating'])))

test_users = [1, 65, 444, 321]
# recommends 4 undiscovered items per each user
results = svd.recommend(test_users, N=4)

# outputs 5 most similar users to user with userId 65
similars = svd.topN_similar(x=65, N=5, column='user')