FIX: fix layer error in GCMC (issue 1631)

recbole/config/configurator.py

@@ -198,7 +198,6 @@ def _merge_external_config_dict(self):
         self.external_config_dict = external_config_dict
 
     def _get_model_and_dataset(self, model, dataset):
-
         if model is None:
             try:
                 model = self.external_config_dict["model"]

recbole/data/dataset/dataset.py

@@ -725,7 +725,6 @@ def _discretization(self):
         dis_info = {}
 
         if self.config["discretization"]:
-
             dis_info = self.config["discretization"]
 
             for field in dis_info.keys():

recbole/evaluator/collector.py

@@ -149,15 +149,13 @@ def eval_batch_collect(
             positive_i(Torch.Tensor): the positive item id for each user.
         """
         if self.register.need("rec.items"):
-
             # get topk
             _, topk_idx = torch.topk(
                 scores_tensor, max(self.topk), dim=-1
             )  # n_users x k
             self.data_struct.update_tensor("rec.items", topk_idx)
 
         if self.register.need("rec.topk"):
-
             _, topk_idx = torch.topk(
                 scores_tensor, max(self.topk), dim=-1
             )  # n_users x k
@@ -169,7 +167,6 @@ def eval_batch_collect(
             self.data_struct.update_tensor("rec.topk", result)
 
         if self.register.need("rec.meanrank"):
-
             desc_scores, desc_index = torch.sort(scores_tensor, dim=-1, descending=True)
 
             # get the index of positive items in the ranking list
@@ -188,7 +185,6 @@ def eval_batch_collect(
             self.data_struct.update_tensor("rec.meanrank", result)
 
         if self.register.need("rec.score"):
-
             self.data_struct.update_tensor("rec.score", scores_tensor)
 
         if self.register.need("data.label"):

recbole/evaluator/register.py

@@ -72,7 +72,6 @@ class Register(object):
     """
 
     def __init__(self, config):
-
         self.config = config
         self.metrics = [metric.lower() for metric in self.config["metrics"]]
         self._build_register()

recbole/model/context_aware_recommender/fignn.py

@@ -105,7 +105,6 @@ def __init__(self, config, dataset):
         self.apply(self._init_weights)
 
     def fignn_layer(self, in_feature):
-
         emb_feature = self.att_embedding(in_feature)
         emb_feature = self.dropout_layer(emb_feature)
         # multi-head self-attention network

recbole/model/context_aware_recommender/fm.py

@@ -27,7 +27,6 @@ class FM(ContextRecommender):
     """Factorization Machine considers the second-order interaction with features to predict the final score."""
 
     def __init__(self, config, dataset):
-
         super(FM, self).__init__(config, dataset)
 
         # define layers and loss

recbole/model/general_recommender/dmf.py

@@ -130,7 +130,6 @@ def _init_weights(self, module):
             normal_(module.weight.data, 0, 0.01)
 
     def forward(self, user, item):
-
         user = self.get_user_embedding(user)
 
         # Following lines construct tensor of shape [B,n_users] using the tensor of shape [B,H]

recbole/model/general_recommender/gcmc.py

@@ -432,7 +432,7 @@ def forward(self, user_X, item_X):
         v_hidden = self.dropout(v_hidden)
 
         u_hidden = self.dense_layer_u(u_hidden)
-        v_hidden = self.dense_layer_u(v_hidden)
+        v_hidden = self.dense_layer_v(v_hidden)
 
         u_outputs = self.dense_activate(u_hidden)
         v_outputs = self.dense_activate(v_hidden)

recbole/model/general_recommender/itemknn.py

@@ -82,7 +82,6 @@ def compute_similarity(self, method, block_size=100):
 
         # Compute all similarities using vectorization
         while start_block < end_local:
-
             end_block = min(start_block + block_size, end_local)
             this_block_size = end_block - start_block
 

recbole/model/general_recommender/line.py

@@ -79,7 +79,6 @@ def get_used_ids(self):
         return cur
 
     def sampler(self, key_ids):
-
         key_ids = np.array(key_ids.cpu())
         key_num = len(key_ids)
         total_num = key_num
@@ -121,14 +120,12 @@ def get_user_id_list(self):
         return np.arange(1, self.n_users)
 
     def forward(self, h, t):
-
         h_embedding = self.user_embedding(h)
         t_embedding = self.item_embedding(t)
 
         return torch.sum(h_embedding.mul(t_embedding), dim=1)
 
     def context_forward(self, h, t, field):
-
         if field == "uu":
             h_embedding = self.user_embedding(h)
             t_embedding = self.item_context_embedding(t)
@@ -139,7 +136,6 @@ def context_forward(self, h, t, field):
         return torch.sum(h_embedding.mul(t_embedding), dim=1)
 
     def calculate_loss(self, interaction):
-
         user = interaction[self.USER_ID]
         pos_item = interaction[self.ITEM_ID]
         neg_item = interaction[self.NEG_ITEM_ID]
@@ -178,7 +174,6 @@ def calculate_loss(self, interaction):
             )
 
     def predict(self, interaction):
-
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
 

recbole/model/general_recommender/macridvae.py

@@ -82,7 +82,6 @@ def reparameterize(self, mu, logvar):
             return mu
 
     def forward(self, rating_matrix):
-
         cores = F.normalize(self.k_embedding.weight, dim=1)
         items = F.normalize(self.item_embedding.weight, dim=1)
 
@@ -127,7 +126,6 @@ def forward(self, rating_matrix):
         return logits, mulist, logvarlist
 
     def calculate_loss(self, interaction):
-
         user = interaction[self.USER_ID]
 
         rating_matrix = self.get_rating_matrix(user)
@@ -169,7 +167,6 @@ def reg_loss(self):
         return loss_1 + loss_2 + loss_3
 
     def predict(self, interaction):
-
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
 

recbole/model/general_recommender/multidae.py

@@ -55,7 +55,6 @@ def mlp_layers(self, layer_dims):
         return nn.Sequential(*mlp_modules)
 
     def forward(self, rating_matrix):
-
         h = F.normalize(rating_matrix)
 
         h = F.dropout(h, self.drop_out, training=self.training)
@@ -64,7 +63,6 @@ def forward(self, rating_matrix):
         return self.decoder(h)
 
     def calculate_loss(self, interaction):
-
         user = interaction[self.USER_ID]
 
         rating_matrix = self.get_rating_matrix(user)
@@ -77,7 +75,6 @@ def calculate_loss(self, interaction):
         return ce_loss
 
     def predict(self, interaction):
-
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
 
@@ -88,7 +85,6 @@ def predict(self, interaction):
         return scores[[torch.arange(len(item)).to(self.device), item]]
 
     def full_sort_predict(self, interaction):
-
         user = interaction[self.USER_ID]
 
         rating_matrix = self.get_rating_matrix(user)

recbole/model/general_recommender/multivae.py

@@ -68,7 +68,6 @@ def reparameterize(self, mu, logvar):
             return mu
 
     def forward(self, rating_matrix):
-
         h = F.normalize(rating_matrix)
 
         h = F.dropout(h, self.drop_out, training=self.training)
@@ -83,7 +82,6 @@ def forward(self, rating_matrix):
         return z, mu, logvar
 
     def calculate_loss(self, interaction):
-
         user = interaction[self.USER_ID]
         rating_matrix = self.get_rating_matrix(user)
 
@@ -108,7 +106,6 @@ def calculate_loss(self, interaction):
         return ce_loss + kl_loss
 
     def predict(self, interaction):
-
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
 

recbole/model/general_recommender/ngcf.py

@@ -146,7 +146,6 @@ def get_ego_embeddings(self):
         return ego_embeddings
 
     def forward(self):
-
         A_hat = (
             self.sparse_dropout(self.norm_adj_matrix)
             if self.node_dropout != 0
@@ -195,7 +194,6 @@ def calculate_loss(self, interaction):
         return mf_loss + self.reg_weight * reg_loss
 
     def predict(self, interaction):
-
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
 

recbole/model/general_recommender/ract.py

@@ -103,7 +103,6 @@ def reparameterize(self, mu, logvar):
             return mu
 
     def forward(self, rating_matrix):
-
         t = F.normalize(rating_matrix)
 
         h = F.dropout(t, self.drop_out, training=self.training) * (1 - self.drop_out)
@@ -127,7 +126,6 @@ def forward(self, rating_matrix):
         return z, mu, logvar
 
     def calculate_actor_loss(self, interaction):
-
         user = interaction[self.USER_ID]
         rating_matrix = self.get_rating_matrix(user)
 
@@ -211,7 +209,6 @@ def calculate_ac_loss(self, interaction):
         return -1 * y
 
     def calculate_loss(self, interaction):
-
         # actor_pretrain
         if self.train_stage == "actor_pretrain":
             return self.calculate_actor_loss(interaction).mean()
@@ -223,7 +220,6 @@ def calculate_loss(self, interaction):
             return self.calculate_ac_loss(interaction).mean()
 
     def predict(self, interaction):
-
         user = interaction[self.USER_ID]
         item = interaction[self.ITEM_ID]
 

recbole/model/knowledge_aware_recommender/mcclk.py

@@ -38,7 +38,6 @@ def __init__(self, item_only=False, attention=True):
     def forward(
         self, entity_emb, user_emb, relation_emb, edge_index, edge_type, inter_matrix
     ):
-
         from torch_scatter import scatter_softmax, scatter_mean
 
         n_entities = entity_emb.shape[0]
@@ -169,7 +168,6 @@ def edge_sampling(self, edge_index, edge_type, rate=0.5):
         return edge_index[:, random_indices], edge_type[random_indices]
 
     def forward(self, user_emb, entity_emb):
-
         # node dropout
         if self.node_dropout_rate > 0.0:
             edge_index, edge_type = self.edge_sampling(
@@ -263,7 +261,6 @@ def build_adj(self, context, topk):
         return L_norm
 
     def _build_graph_separately(self, entity_emb):
-
         # node dropout
         if self.node_dropout_rate > 0.0:
             edge_index, edge_type = self.edge_sampling(
@@ -455,7 +452,6 @@ def get_edges(self, graph):
         return index.to(self.device), type.to(self.device)
 
     def forward(self):
-
         user_emb = self.user_embedding.weight
         entity_emb = self.entity_embedding.weight
         # Construct a k-Nearest-Neighbor item-item semantic graph and Structural View Encoder
@@ -503,7 +499,6 @@ def sim(self, z1: torch.Tensor, z2: torch.Tensor):
         return torch.mm(z1, z2.t())
 
     def calculate_loss(self, interaction):
-
         if self.restore_user_e is not None or self.restore_item_e is not None:
             self.restore_user_e, self.restore_item_e = None, None
 

recbole/model/layers.py

@@ -596,7 +596,6 @@ def __init__(
         hidden_act="gelu",
         layer_norm_eps=1e-12,
     ):
-
         super(TransformerEncoder, self).__init__()
         layer = TransformerLayer(
             n_heads,
@@ -825,7 +824,6 @@ def __init__(
         hidden_act="gelu",
         layer_norm_eps=1e-12,
     ):
-
         super(LightTransformerEncoder, self).__init__()
         layer = LightTransformerLayer(
             n_heads,
@@ -1354,7 +1352,6 @@ class FMFirstOrderLinear(nn.Module):
     """
 
     def __init__(self, config, dataset, output_dim=1):
-
         super(FMFirstOrderLinear, self).__init__()
         self.field_names = dataset.fields(
             source=[

recbole/model/sequential_recommender/dien.py

@@ -123,7 +123,6 @@ def _init_weights(self, module):
                 constant_(module.bias.data, 0)
 
     def forward(self, user, item_seq, neg_item_seq, item_seq_len, next_items):
-
         max_length = item_seq.shape[1]
         # concatenate the history item seq with the target item to get embedding together
         item_seq_next_item = torch.cat(

recbole/model/sequential_recommender/din.py

@@ -110,7 +110,6 @@ def _init_weights(self, module):
                 constant_(module.bias.data, 0)
 
     def forward(self, user, item_seq, item_seq_len, next_items):
-
         max_length = item_seq.shape[1]
         # concatenate the history item seq with the target item to get embedding together
         item_seq_next_item = torch.cat((item_seq, next_items.unsqueeze(1)), dim=-1)

recbole/model/sequential_recommender/fossil.py

@@ -92,7 +92,6 @@ def inverse_seq_item_embedding(self, seq_item_embedding, seq_item_len):
         return short_item_embedding
 
     def reg_loss(self, user_embedding, item_embedding, seq_output):
-
         reg_1 = self.reg_weight
         loss_1 = (
             reg_1 * torch.norm(user_embedding, p=2)
@@ -103,12 +102,10 @@ def reg_loss(self, user_embedding, item_embedding, seq_output):
         return loss_1
 
     def init_weights(self, module):
-
         if isinstance(module, nn.Embedding) or isinstance(module, nn.Linear):
             xavier_normal_(module.weight.data)
 
     def forward(self, seq_item, seq_item_len, user):
-
         seq_item_embedding = self.item_embedding(seq_item)
 
         high_order_seq_item_embedding = self.inverse_seq_item_embedding(
@@ -152,7 +149,6 @@ def get_similarity(self, seq_item_embedding, seq_item_len):
         return similarity
 
     def calculate_loss(self, interaction):
-
         seq_item = interaction[self.ITEM_SEQ]
         user = interaction[self.USER_ID]
         seq_item_len = interaction[self.ITEM_SEQ_LEN]
@@ -176,7 +172,6 @@ def calculate_loss(self, interaction):
             return loss + self.reg_loss(user_lambda, pos_items_embedding, seq_output)
 
     def predict(self, interaction):
-
         item_seq = interaction[self.ITEM_SEQ]
         item_seq_len = interaction[self.ITEM_SEQ_LEN]
         test_item = interaction[self.ITEM_ID]
@@ -187,7 +182,6 @@ def predict(self, interaction):
         return scores
 
     def full_sort_predict(self, interaction):
-
         item_seq = interaction[self.ITEM_SEQ]
         user = interaction[self.USER_ID]
         item_seq_len = interaction[self.ITEM_SEQ_LEN]