from DRecPy.Recommender.Baseline import ItemKNN
from DRecPy.Dataset import get_full_dataset
from DRecPy.Evaluation.Splits import leave_k_out
from DRecPy.Evaluation.Processes import ranking_evaluation
from DRecPy.Evaluation.Metrics import HitRatio
from DRecPy.Evaluation.Metrics import NDCG
import time
ds_full = get_full_dataset('ml-100k')
ds_train, ds_test = leave_k_out(ds_full, k=1, last_timestamps=True, seed=0)
start_train = time.time()
item_cf = ItemKNN(k=5,
m=1,
shrinkage=50,
sim_metric='adjusted_cosine',
verbose=True)
item_cf.fit(ds_train)
print("Training took", time.time() - start_train)
start_evaluation = time.time()
print(
ranking_evaluation(item_cf,
ds_test,
n_pos_interactions=1,
n_neg_interactions=100,
generate_negative_pairs=True,
novelty=True,
k=list(range(1, 11)),
metrics=[HitRatio(), NDCG()],
seed=10))
def fit_model_mean_aggr(train_interaction_ds):
fit_model_mean_aggr = ItemKNN(k=20, m=5, sim_metric='adjusted_cosine', aggregation='mean', shrinkage=100,
use_averages=False)
fit_model_mean_aggr.fit(train_interaction_ds)
return fit_model_mean_aggr
def fit_model_no_shrinkage(train_interaction_ds):
fit_model_no_shrinkage = ItemKNN(k=20, m=5, sim_metric='adjusted_cosine', aggregation='weighted_mean',
shrinkage=None, use_averages=False)
fit_model_no_shrinkage.fit(train_interaction_ds)
return fit_model_no_shrinkage
def fit_model_cosine_sim(train_interaction_ds):
fit_model_cosine_sim = ItemKNN(k=20, m=5, sim_metric='cosine', aggregation='weighted_mean', shrinkage=100,
use_averages=False)
fit_model_cosine_sim.fit(train_interaction_ds)
return fit_model_cosine_sim
def fit_model_use_averages(train_interaction_ds):
fit_model_use_averages = ItemKNN(k=1, m=1, sim_metric='adjusted_cosine', aggregation='weighted_mean', shrinkage=100,
use_averages=True)
fit_model_use_averages.fit(train_interaction_ds)
return fit_model_use_averages