Fix item centering. Add item-specific L2 regularization.

recbole/model/general_recommender/admmslim.py

@@ -44,21 +44,28 @@ def __init__(self, config, dataset):
 
         lambda1 = config['lambda1']
         lambda2 = config['lambda2']
+        alpha = config['alpha']
         rho = config['rho']
         k = config['k']
         positive_only = config['positive_only']
         self.center_columns = config['center_columns']
+        self.item_means = X.mean(axis=0).getA1()
 
         if self.center_columns:
-            self.item_means = X.mean(axis=0).getA1()
             zero_mean_X = X.toarray() - self.item_means
             G = (zero_mean_X.T @ zero_mean_X)
             # large memory cost because we need to make X dense to subtract mean, delete asap
             del zero_mean_X
         else:
             G = (X.T @ X).toarray()
 
-        diag = (lambda2 + rho) * np.identity(num_items)
+
+        # alpha = 0 corresponds to this case
+        # diag = (lambda2 + rho) * np.identity(num_items)
+
+        # add 1 to means to prevent singular matrix in what follows
+        diag = (lambda2 + rho) * np.diag(np.power(self.item_means + 1, alpha / 2))
+
         P = np.linalg.inv(G + diag)
         B_aux = P @ G
 
@@ -93,22 +100,20 @@ def predict(self, interaction):
         user_interactions = self.interaction_matrix[user, :].toarray()
 
         if self.center_columns:
-            user_interactions -= self.item_means
-
-        r = torch.from_numpy(
-            (user_interactions * self.item_similarity[:, item].T).sum(axis=1).flatten())
+            r = (((user_interactions - self.item_means) * self.item_similarity[:, item].T).sum(axis=1)).flatten() + self.item_means[item]
+        else:
+            r = (user_interactions * self.item_similarity[:, item].T).sum(axis=1).flatten()
 
-        return add_noise(r)
+        return add_noise(torch.from_numpy(r))
 
     def full_sort_predict(self, interaction):
         user = interaction[self.USER_ID].cpu().numpy()
 
         user_interactions = self.interaction_matrix[user, :].toarray()
 
         if self.center_columns:
-            user_interactions -= self.item_means
-
-        r = user_interactions @ self.item_similarity
-        r = torch.from_numpy(r.flatten())
+            r = ((user_interactions - self.item_means) @ self.item_similarity + self.item_means).flatten()
+        else:
+            r = (user_interactions @ self.item_similarity).flatten()
 
-        return add_noise(r)
+        return add_noise(torch.from_numpy(r))

recbole/properties/model/ADMMSLIM.yaml

@@ -1,6 +1,7 @@
 lambda1: 3
-lambda2: 150
-rho: 300
+lambda2: 200
+alpha: 0.25
+rho: 4000
 k: 100
-positive_only: False
+positive_only: True
 center_columns: False