updates with more comments and descriptions

StanfordASL · Aug 3, 2021 · 883aa8f · 883aa8f
2 parents b60d4bd + ce882d8
commit 883aa8f
Showing 1 changed file with 5 additions and 5 deletions.
diff --git a/src/stlcg.py b/src/stlcg.py
@@ -744,8 +744,8 @@ def robustness_trace(self, inputs, pscale=1, scale=-1, keepdim=True, agm=False,
         minish = Minish()
         maxish = Maxish()
         LHS = trace2.unsqueeze(-1).repeat([1, 1, 1,trace2.shape[1]]).permute(0, 3, 2, 1)
-        RHS = torch.ones_like(LHS)*-LARGE_NUMBER
         if interval == None:
+            RHS = torch.ones_like(LHS)*-LARGE_NUMBER
             for i in range(trace2.shape[1]):
                 RHS[:,i:,:,i] = Alw(trace1[:,i:,:])
             return maxish(
@@ -758,7 +758,7 @@ def robustness_trace(self, inputs, pscale=1, scale=-1, keepdim=True, agm=False,
             for i in range(b,trace2.shape[1]):
                 A = trace2[:,i-b:i-a+1,:].unsqueeze(-1)
                 relevant = trace1[:,:i+1,:]
-                B = Alw(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:b+1,:].unsqueeze(-1)
+                B = Alw(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:b+1,:].flip(1).unsqueeze(-1)
                 RHS.append(maxish(minish(torch.cat([A,B], dim=-1), dim=-1, scale=scale, keepdim=False, distributed=distributed), dim=1, scale=scale, keepdim=keepdim, distributed=distributed))
             return torch.cat(RHS, dim=1);
         else:
@@ -767,7 +767,7 @@ def robustness_trace(self, inputs, pscale=1, scale=-1, keepdim=True, agm=False,
             for i in range(a,trace2.shape[1]):
                 A = trace2[:,:i-a+1,:].unsqueeze(-1)
                 relevant = trace1[:,:i+1,:]
-                B = Alw(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:,:].unsqueeze(-1)
+                B = Alw(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:,:].flip(1).unsqueeze(-1)
                 RHS.append(maxish(minish(torch.cat([A,B], dim=-1), dim=-1, scale=scale, keepdim=False, distributed=distributed), dim=1, scale=scale, keepdim=keepdim, distributed=distributed))
             return torch.cat(RHS, dim=1);
 
@@ -823,7 +823,7 @@ def robustness_trace(self, inputs, pscale=1, scale=-1, keepdim=True, agm=False,
             for i in range(b,trace2.shape[1]):
                 A = trace2[:,i-b:i-a+1,:].unsqueeze(-1)
                 relevant = trace1[:,:i+1,:]
-                B = Ev(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:b+1,:].unsqueeze(-1)
+                B = Ev(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:b+1,:].flip(1).unsqueeze(-1)
                 RHS.append(maxish(minish(torch.cat([A,B], dim=-1), dim=-1, scale=scale, keepdim=False, distributed=distributed), dim=1, scale=scale, keepdim=keepdim, distributed=distributed))
             return torch.cat(RHS, dim=1);
         else:
@@ -832,7 +832,7 @@ def robustness_trace(self, inputs, pscale=1, scale=-1, keepdim=True, agm=False,
             for i in range(a,trace2.shape[1]):
                 A = trace2[:,:i-a+1,:].unsqueeze(-1)
                 relevant = trace1[:,:i+1,:]
-                B = Ev(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:,:].unsqueeze(-1)
+                B = Ev(relevant.flip(1), scale=scale, keepdim=keepdim, distributed=distributed)[:,a:,:].flip(1).unsqueeze(-1)
                 RHS.append(maxish(minish(torch.cat([A,B], dim=-1), dim=-1, scale=scale, keepdim=False, distributed=distributed), dim=1, scale=scale, keepdim=keepdim, distributed=distributed))
             return torch.cat(RHS, dim=1);                                                    # [batch_size, time_dim, x_dim]