def test_mean_center_predictor_algo(self):
algo = UserBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="person"),
topk=2,
predictor_config=PredictorConfig(name="mean_center")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.626923301909203
def test_item_based_regression_model(self):
algo = ItemBasedRegressionModel(
RegressionModelNeighborhoodBasedConfig(topk=3,
lr=0.005,
epochs=200,
wdecay=0.0001))
algo.fit(test_rating_data)
hat_rating = algo.predict()
print(statistic(ma.masked_equal(test_rating_data, 0), hat_rating))
print(rmse(ma.masked_equal(test_rating_data, 0), hat_rating))
print(algo.__parameters__())
def test_idf_sim_algo(self):
algo = UserBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="idf_person"),
topk=2,
sim_threshold=None,
predictor_config=PredictorConfig(name="mean_center")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.9546439193233666
def test_pca_person_sim_algo(self):
algo = ItemBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="pca_person", dims=2),
topk=2,
sim_threshold=None,
predictor_config=PredictorConfig(name="norm")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.6237294778495414
def test_amplify_sim_algo(self):
algo = UserBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="amplify_person",
amplify_alpha=3),
topk=2,
sim_threshold=None,
predictor_config=PredictorConfig(name="mean_sigma")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.5921262449508355
def test_mean_sigma_predictor_algo(self):
algo = UserBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="discounted_person",
discounted_beta=6),
topk=2,
sim_threshold=None,
predictor_config=PredictorConfig(name="mean_sigma")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.6249970642035232
def test_norm_predictor_algo(self):
algo = UserBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="person",
discounted_beta=None),
topk=2,
sim_threshold=None,
predictor_config=PredictorConfig(name="norm")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 2.125021074648673
def test_mean_center_predictor_algo_with_sim_threshold(self):
algo = UserBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="person",
discounted_beta=None),
topk=None,
sim_threshold=0.0,
predictor_config=PredictorConfig(name="mean_center")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.6314698247230878
def test_mean_center_predictor_algo(self):
algo = ItemBasedNeighborhoodAlgorithm(
NeighborhoodBasedConfig(
sim_config=SimilarityConfig(name="cosine"),
topk=2,
sim_threshold=None,
predictor_config=PredictorConfig(name="mean_center")))
algo.fit(test_rating_data)
rating_hat = algo.predict(None)
print(rating_hat)
print(rmse(rating_hat, ma.masked_equal(test_rating_data,
0))) # 0.6543287113746402
parser.add_argument("--epochs", type=int, default=10000, help="epochs about regression model")
parser.add_argument("--weight_decay", type=float, default=0.001, help="weight decay or l2 regularization")
parser.add_argument("--check_gradient", type=bool, default=False, help="check gradient in optimization")
parser.add_argument("--model_dir", type=str, help="model directory to save and restore")
if __name__ == "__main__":
args = parser.parse_args()
logger.info("--------------------------------------------------")
print("neighborhood args: {}".format(args))
ml = MovieLenDataset(args.dataset)
if args.algo == "user":
algorithm = UserBasedNeighborhoodAlgorithm(NeighborhoodBasedConfig(sim_config=SimilarityConfig(name=args.sim, discounted_beta=args.discounted_beta, amplify_alpha=args.amplify_alpha, dims=args.dims), topk=args.topk, sim_threshold=args.sim_threshold, predictor_config=PredictorConfig(name=args.predictor)))
elif args.algo == "item":
algorithm = ItemBasedNeighborhoodAlgorithm(NeighborhoodBasedConfig(sim_config=SimilarityConfig(name=args.sim, discounted_beta=args.discounted_beta, amplify_alpha=args.amplify_alpha, dims=args.dims), topk=args.topk, sim_threshold=args.sim_threshold, predictor_config=PredictorConfig(name=args.predictor)))
elif args.algo == "user_reg":
algorithm = UserBasedRegressionModel(RegressionModelNeighborhoodBasedConfig(topk=args.topk, lr=args.learn_rate, epochs=args.epochs, wdecay=args.weight_decay, check_gradient=args.check_gradient, model_dir=args.model_dir))
elif args.algo == "item_reg":
algorithm = ItemBasedRegressionModel(RegressionModelNeighborhoodBasedConfig(topk=args.topk, lr=args.learn_rate, epochs=args.epochs, wdecay=args.weight_decay, check_gradient=args.check_gradient, model_dir=args.model_dir))
else:
print("[USAGE] algo must be in [user,item,user_reg,item_reg]")
sys.exit(-1)
algorithm.fit(ml.R)
hat_rating = algorithm.predict(None)
rating = ma.masked_equal(ml.R, 0)
logger.debug("rating_count: {rating_count}\tpredict_count:{predict_count}\tpredict_rate:{predict_rate}".format(**metric.statistic(rating, hat_rating)))
logger.info("|{}|{}|".format(args, metric.rmse(rating[:, 0], hat_rating[:, 0])))
parser.add_argument("--epochs",
type=int,
default=1000,
help="epochs about regression model")
parser.add_argument("--model_dir",
type=str,
help="model directory to save and restore")
if __name__ == "__main__":
args = parser.parse_args()
logger.info("--------------------------------------------------")
print("args: {}".format(args))
ml = MovieLenDataset(args.dataset)
if args.algo == "umf":
config = LatentFactorConfig(model_dir=args.model_dir,
embedding_size=args.embedding_size,
epochs=args.epochs,
learn_rate=args.learn_rate)
algorithm = UnconstrainedMatrixFactorAlgorithm(config)
else:
print("[USAGE] algo must be in [umf, ]")
sys.exit(-1)
algorithm.fit(ml.R)
hat_rating = algorithm.predict(None)
rating = ma.masked_equal(ml.R, 0)
logger.info("|{}|{}|".format(args,
metric.rmse(rating[:, 0], hat_rating[:, 0])))
def test_rmse(self):
y = ma.masked_equal([1, 2, 0, 1], 0)
y_hat = ma.masked_equal([0, 1, 2, 2], 0)
assert math.isclose(metric.rmse(y, y_hat), 1)
