bugfix - versions with and without bp_to_u working

spline_conv.py

@@ -31,10 +31,16 @@ def spline_conv(
     # Convert to [|E| x M_in] feature matrix and calculate [|E| x M_out].
 
     if output.is_cuda:
-        output = SplineConvGPU(kernel_size, is_open_spline, K, degree,
-                               basis_kernel, basis_backward_kernel,
-                               weighting_kernel, weighting_backward_kernel,
-                               bp_to_adj)(output, weight[:-1], values)
+        if bp_to_adj:
+            output = SplineConvGPU(kernel_size, is_open_spline, K, degree,
+                                   basis_kernel, basis_backward_kernel,
+                                   weighting_kernel, weighting_backward_kernel,
+                                   bp_to_adj)(output, weight[:-1], values)
+        else:
+            output = SplineConvGPU(kernel_size, is_open_spline, K, degree,
+                                   basis_kernel, basis_backward_kernel,
+                                   weighting_kernel, weighting_backward_kernel,
+                                   bp_to_adj, values)(output, weight[:-1])
     else:
         # CPU Implementation not available
         raise NotImplementedError()
@@ -50,8 +56,8 @@ def spline_conv(
     output += torch.mm(input, weight[-1])
 
     # Normalize output by degree.
-    ones = values.data.new(values.size(0)).fill_(1)
-    zero = values.data.new(output.size(0)).fill_(0)
+    ones = output.data.new(values.size(0)).fill_(1)
+    zero = output.data.new(output.size(0)).fill_(0)
     degree = zero.scatter_add_(0, row, ones)
     degree = torch.clamp(degree, min=1)
     output = output / Variable(degree.view(-1, 1))

spline_conv_gpu.py

@@ -471,7 +471,7 @@ class SplineConvGPU(Function):
     def __init__(self, kernel_size, is_open_spline, K, degree,
                                basis_kernel, basis_backward_kernel,
                                weighting_kernel, weighting_backward_kernel,
-                               bp_to_adj=False):
+                               bp_to_adj=False, adj_values=None):
         super(SplineConvGPU, self).__init__()
         self.degree = degree
         self.f_weighting_fw = weighting_kernel
@@ -481,11 +481,15 @@ class SplineConvGPU(Function):
         self.f_basis_fw = basis_kernel
         self.f_basis_bw = basis_backward_kernel
         self.bp_to_adj = bp_to_adj
+        self.adj_values = adj_values
 
-    def forward(self, input, weight, adj_values):
+    def forward(self, input, weight, adj_values=None):
         assert input.is_cuda and weight.is_cuda
         self.K, self.M_in, self.M_out = weight.size()
 
+        # If bp_to_u is false
+        if adj_values is None:
+            adj_values = self.adj_values
         # Compute B-spline basis tensor products
         adj_values = adj_values.unsqueeze(1) if len(adj_values.size()) < 2 \
             else adj_values