FIX: fix bugs when running Random, ADMMSLIM, and SLIMElastic with negative sampling evaluations

recbole/config/configurator.py

@@ -433,9 +433,9 @@ def _set_default_parameters(self):
                 )
             for op_args in default_train_neg_sample_args:
                 if op_args not in self.final_config_dict["train_neg_sample_args"]:
-                    self.final_config_dict["train_neg_sample_args"][
-                        op_args
-                    ] = default_train_neg_sample_args[op_args]
+                    self.final_config_dict["train_neg_sample_args"][op_args] = (
+                        default_train_neg_sample_args[op_args]
+                    )
 
         # eval_args checking
         default_eval_args = {

recbole/data/dataset/customized_dataset.py

@@ -146,9 +146,9 @@ def data_augmentation(self):
                     for i, (index, length) in enumerate(
                         zip(item_list_index, item_list_length)
                     ):
-                        new_dict[self.neg_item_list_field][i][
-                            :length
-                        ] = self.neg_item_list[index]
+                        new_dict[self.neg_item_list_field][i][:length] = (
+                            self.neg_item_list[index]
+                        )
 
         new_data.update(Interaction(new_dict))
         self.inter_feat = new_data

recbole/data/dataset/dataset.py

@@ -651,9 +651,9 @@ def _fill_nan(self):
                 else:
                     dtype = np.int64 if ftype == FeatureType.TOKEN_SEQ else np.float
                     feat[field] = feat[field].apply(
-                        lambda x: np.array([], dtype=dtype)
-                        if isinstance(x, float)
-                        else x
+                        lambda x: (
+                            np.array([], dtype=dtype) if isinstance(x, float) else x
+                        )
                     )
 
     def _normalize(self):

recbole/data/transform.py

@@ -289,9 +289,9 @@ def __call__(self, dataset, interaction):
 
             shuffle_index = list(range(reorder_begin, reorder_begin + reorder_len))
             random.shuffle(shuffle_index)
-            reorder_item_seq[
-                reorder_begin : reorder_begin + reorder_len
-            ] = reorder_item_seq[shuffle_index]
+            reorder_item_seq[reorder_begin : reorder_begin + reorder_len] = (
+                reorder_item_seq[shuffle_index]
+            )
 
             reorder_seq_list.append(
                 torch.tensor(reorder_item_seq, dtype=torch.long, device=device)

recbole/evaluator/metrics.py

@@ -253,6 +253,7 @@ class GAUC(AbstractMetric):
 
     :math:`rank_i` is the descending rank of the i-th items in :math:`R(u)`.
     """
+
     metric_type = EvaluatorType.RANKING
     metric_need = ["rec.meanrank"]
 
@@ -393,6 +394,7 @@ class MAE(LossMetric):
 
     :math:`|S|` represents the number of pairs in :math:`S`.
     """
+
     smaller = True
 
     def __init__(self, config):
@@ -413,6 +415,7 @@ class RMSE(LossMetric):
     .. math::
        \mathrm{RMSE} = \sqrt{\frac{1}{|{S}|} \sum_{(u, i) \in {S}}(\hat{r}_{u i}-r_{u i})^{2}}
     """
+
     smaller = True
 
     def __init__(self, config):
@@ -434,6 +437,7 @@ class LogLoss(LossMetric):
     .. math::
         LogLoss = \frac{1}{|S|} \sum_{(u,i) \in S}(-((r_{u i} \ \log{\hat{r}_{u i}}) + {(1 - r_{u i})}\ \log{(1 - \hat{r}_{u i})}))
     """
+
     smaller = True
 
     def __init__(self, config):
@@ -461,6 +465,7 @@ class ItemCoverage(AbstractMetric):
     .. math::
        \mathrm{Coverage@K}=\frac{\left| \bigcup_{u \in U} \hat{R}(u) \right|}{|I|}
     """
+
     metric_type = EvaluatorType.RANKING
     metric_need = ["rec.items", "data.num_items"]
 
@@ -509,6 +514,7 @@ class AveragePopularity(AbstractMetric):
 
     :math:`\phi(i)` is the number of interaction of item i in training data.
     """
+
     metric_type = EvaluatorType.RANKING
     smaller = True
     metric_need = ["rec.items", "data.count_items"]
@@ -582,6 +588,7 @@ class ShannonEntropy(AbstractMetric):
     :math:`p(i)` is the probability of recommending item i
     which is the number of item i in recommended list over all items.
     """
+
     metric_type = EvaluatorType.RANKING
     metric_need = ["rec.items"]
 
@@ -637,6 +644,7 @@ class GiniIndex(AbstractMetric):
     :math:`P{(i)}` represents the number of times all items appearing in the recommended list,
     which is indexed in non-decreasing order (P_{(i)} \leq P_{(i+1)}).
     """
+
     metric_type = EvaluatorType.RANKING
     smaller = True
     metric_need = ["rec.items", "data.num_items"]
@@ -699,6 +707,7 @@ class TailPercentage(AbstractMetric):
         If you want to use this metric, please set the parameter 'tail_ratio' in the config
         which can be an integer or a float in (0,1]. Otherwise it will default to 0.1.
     """
+
     metric_type = EvaluatorType.RANKING
     metric_need = ["rec.items", "data.count_items"]
 

recbole/model/exlib_recommender/lightgbm.py

@@ -14,6 +14,7 @@
 
 class LightGBM(lgb.Booster):
     r"""LightGBM is inherited from lgb.Booster"""
+
     type = ModelType.DECISIONTREE
     input_type = InputType.POINTWISE
 

recbole/model/exlib_recommender/xgboost.py

@@ -14,6 +14,7 @@
 
 class XGBoost(xgb.Booster):
     r"""XGBoost is inherited from xgb.Booster"""
+
     type = ModelType.DECISIONTREE
     input_type = InputType.POINTWISE
 

recbole/model/general_recommender/admmslim.py

@@ -111,7 +111,7 @@ def predict(self, interaction):
                 .flatten()
             )
 
-        return add_noise(torch.from_numpy(r))
+        return add_noise(torch.from_numpy(r)).to(self.device)
 
     def full_sort_predict(self, interaction):
         user = interaction[self.USER_ID].cpu().numpy()

recbole/model/general_recommender/bpr.py

@@ -26,6 +26,7 @@
 
 class BPR(GeneralRecommender):
     r"""BPR is a basic matrix factorization model that be trained in the pairwise way."""
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/cdae.py

@@ -26,6 +26,7 @@ class CDAE(GeneralRecommender, AutoEncoderMixin):
     for top-N recommendation that utilizes the idea of Denoising Auto-Encoders.
     We implement the the CDAE model with only user dataloader.
     """
+
     input_type = InputType.POINTWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/convncf.py

@@ -56,6 +56,7 @@ class ConvNCF(GeneralRecommender):
     We carefully design the data interface and use sparse tensor to train and test efficiently.
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/dgcf.py

@@ -59,6 +59,7 @@ class DGCF(GeneralRecommender):
     we carefully design the data interface and use sparse tensor to train and test efficiently.
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/diffrec.py

@@ -86,6 +86,7 @@ class DiffRec(GeneralRecommender, AutoEncoderMixin):
     Note that DiffRec simultaneously ranks all items for each user.
     We implement the the DiffRec model with only user dataloader.
     """
+
     input_type = InputType.LISTWISE
 
     def __init__(self, config, dataset):
@@ -141,9 +142,9 @@ def __init__(self, config, dataset):
                 self.device
             )
             if self.beta_fixed:
-                self.betas[
-                    0
-                ] = 0.00001  # Deep Unsupervised Learning using Noneequilibrium Thermodynamics 2.4.1
+                self.betas[0] = (
+                    0.00001  # Deep Unsupervised Learning using Noneequilibrium Thermodynamics 2.4.1
+                )
                 # The variance \beta_1 of the first step is fixed to a small constant to prevent overfitting.
             assert len(self.betas.shape) == 1, "betas must be 1-D"
             assert (

recbole/model/general_recommender/dmf.py

@@ -37,6 +37,7 @@ class DMF(GeneralRecommender):
         For a better performance and stability, we replace cosine similarity to inner-product when calculate
         final score of user's and item's embedding.
     """
+
     input_type = InputType.POINTWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/ease.py

@@ -18,6 +18,7 @@ class EASE(GeneralRecommender):
     strengths of auto-encoders and neighborhood-based approaches.
 
     """
+
     input_type = InputType.POINTWISE
     type = ModelType.TRADITIONAL
 

recbole/model/general_recommender/gcmc.py

@@ -48,6 +48,7 @@ class GCMC(GeneralRecommender):
 
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/itemknn.py

@@ -161,6 +161,7 @@ def compute_similarity(self, method, block_size=100):
 
 class ItemKNN(GeneralRecommender):
     r"""ItemKNN is a basic model that compute item similarity with the interaction matrix."""
+
     input_type = InputType.POINTWISE
     type = ModelType.TRADITIONAL
 

recbole/model/general_recommender/lightgcn.py

@@ -39,6 +39,7 @@ class LightGCN(GeneralRecommender):
 
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/line.py

@@ -37,6 +37,7 @@ class LINE(GeneralRecommender):
 
     We implement the model to train users and items embedding for recommendation.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/macridvae.py

@@ -34,6 +34,7 @@ class MacridVAE(GeneralRecommender, AutoEncoderMixin):
 
     We implement the model following the original author.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/multidae.py

@@ -26,6 +26,7 @@ class MultiDAE(GeneralRecommender, AutoEncoderMixin):
 
     We implement the the MultiDAE model with only user dataloader.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/multivae.py

@@ -25,6 +25,7 @@ class MultiVAE(GeneralRecommender, AutoEncoderMixin):
 
     We implement the MultiVAE model with only user dataloader.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/ncl.py

@@ -23,6 +23,7 @@ class NCL(GeneralRecommender):
     r"""NCL is a neighborhood-enriched contrastive learning paradigm for graph collaborative filtering.
     Both structural and semantic neighbors are explicitly captured as contrastive learning objects.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/neumf.py

@@ -33,6 +33,7 @@ class NeuMF(GeneralRecommender):
         Our implementation only contains a rough pretraining function.
 
     """
+
     input_type = InputType.POINTWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/ngcf.py

@@ -36,6 +36,7 @@ class NGCF(GeneralRecommender):
     r"""NGCF is a model that incorporate GNN for recommendation.
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/nncf.py

@@ -30,6 +30,7 @@ class NNCF(GeneralRecommender):
     r"""NNCF is an neural network enhanced matrix factorization model which also captures neighborhood information.
     We implement the NNCF model with three ways to process neighborhood information.
     """
+
     input_type = InputType.POINTWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/pop.py

@@ -25,6 +25,7 @@
 
 class Pop(GeneralRecommender):
     r"""Pop is an fundamental model that always recommend the most popular item."""
+
     input_type = InputType.POINTWISE
     type = ModelType.TRADITIONAL
 

recbole/model/general_recommender/ract.py

@@ -26,6 +26,7 @@ class RaCT(GeneralRecommender, AutoEncoderMixin):
 
     We implement the RaCT model with only user dataloader.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/random.py

@@ -35,7 +35,7 @@ def calculate_loss(self, interaction):
         return torch.nn.Parameter(torch.zeros(1))
 
     def predict(self, interaction):
-        return torch.rand(len(interaction)).squeeze(-1)
+        return torch.rand(len(interaction), device=self.device).squeeze(-1)
 
     def full_sort_predict(self, interaction):
         batch_user_num = interaction[self.USER_ID].shape[0]

recbole/model/general_recommender/recvae.py

@@ -110,6 +110,7 @@ class RecVAE(GeneralRecommender, AutoEncoderMixin):
 
     We implement the model following the original author
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/sgl.py

@@ -35,6 +35,7 @@ class SGL(GeneralRecommender):
 
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/simplex.py

@@ -34,6 +34,7 @@ class SimpleX(GeneralRecommender):
 
     We implement the model following the original author with a pairwise training mode.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):

recbole/model/general_recommender/slimelastic.py

@@ -26,6 +26,7 @@ class SLIMElastic(GeneralRecommender):
     regularized optimization problem.
 
     """
+
     input_type = InputType.POINTWISE
     type = ModelType.TRADITIONAL
 
@@ -99,7 +100,7 @@ def predict(self, interaction):
             (self.interaction_matrix[user, :].multiply(self.item_similarity[:, item].T))
             .sum(axis=1)
             .getA1()
-        )
+        ).to(self.device)
 
         return r
 

recbole/model/general_recommender/spectralcf.py

@@ -46,6 +46,7 @@ class SpectralCF(GeneralRecommender):
         For a better stability, we replace :math:`U U^T` with identity matrix :math:`I` and
         replace :math:`U \Lambda U^T` with laplace matrix :math:`L`.
     """
+
     input_type = InputType.PAIRWISE
 
     def __init__(self, config, dataset):