Update modeling_super_linear.py

modeling_super_linear.py

@@ -200,40 +200,22 @@ class RLinear(nn.Module):
         self.zero_shot_Linear    = None
 
     def transform_model(self,new_lookback,mode):
-        if mode == 2: 
+        if mode == 1:
             W              = self.Linear.weight.detach()
             new_W          = W[:, -new_lookback:] 
             original_norm  = torch.norm(W, p=2)
             new_norm       = torch.norm(new_W, p=2)
             final_scaling  = original_norm / new_norm if new_norm.item() != 0 else 1.0
-            final_scaling  = 1
             new_W          = new_W * final_scaling
-            self.zero_shot_Linear        = new_W
 
+            self.zero_shot_Linear        = new_W
         else:
-            W = self.Linear.weight.detach()
-            target_indices = torch.linspace(0, self.seq_len - 1, steps=new_lookback, device=W.device)
-            source_indices = torch.arange(0, self.seq_len, device=W.device).float()
-
-            # Initialize the new weight matrix
-            new_W = torch.zeros((W.size(0), new_lookback), device=W.device)
-
-            # Linear interpolation for each row
-            for i in range(W.size(0)):
-                new_W[i] = torch.tensor([torch.sum(W[i] * (1 - torch.abs(idx - source_indices) / self.seq_len).clamp(min=0)) 
-                                        for idx in target_indices], device=W.device)
-
-            # Maintain the same norm as the original weights
-            original_norm = torch.norm(W, p=2)
-            new_norm = torch.norm(new_W, p=2)
-            final_scaling = original_norm / new_norm if new_norm.item() != 0 else 1.0
-            new_W = new_W * final_scaling
+            W              = self.Linear.weight.detach()
+            new_W          =  F.interpolate(W.unsqueeze(0), size=(new_lookback, self.horizon ), mode='bilinear', align_corners=False).squeeze(0)
+            self.zero_shot_Linear        = new_W
 
-            self.zero_shot_Linear = new_W
             
 
-    
-
     def forward(self, x):
         # x: [Batch, Input length,Channel]
         x_shape = x.shape