Pathfinder Discovery Netwrok

[riskiem]

Hi, I am trying to implement the Pathfinder discovery network from the following paper:

https://arxiv.org/pdf/2010.12878.pdf

I have 25 graphs, with a test train split of 21-4. The graphs have 50 nodes each, 1 node feature and 1 edge feature.

My network is as follows:

import torch
from tqdm import tqdm
import torch.nn.functional as F
from torch_geometric.nn import GCNConv

class PathfinderDiscoveryNetwork(torch.nn.Module):
    #def __init__(self,node_filters,edge_filters):
    def __init__(self,node_filters,edge_filters):
        super(PathfinderDiscoveryNetwork, self).__init__()
        self.dense_1 = torch.nn.Linear(data.edge_attr, edge_filters)
        self.dense_2 = torch.nn.Linear(edge_filters, 2)
        self.convolution_1 = GCNConv(data.x, node_filters)
        self.convolution_2 = GCNConv(node_filters, 2)

    def forward(self, x, edge_index, edge_attr):
        edge_attr = F.relu(self.dense_1(edge_attr))
        edge_attr = torch.sigmoid(self.dense_2(edge_attr)).view(-1)
        x = self.convolution_1(x, edge_index, edge_attr)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.convolution_2(x, edge_index, edge_attr)
        return F.log_softmax(x, dim=1)

 model = PathfinderDiscoveryNetwork(32,32)
 optimizer = torch.optim.Adam(model.parameters(), lr=0.01)                                      
 criterion = torch.nn.CrossEntropyLoss()  

def train(epoch):
    model.train()
    loss_all = 0
    num_nodes = 0
    for data in train_loader:
        
        optimizer.zero_grad()
        out = model(data.x, data.edge_index, data.edge_attr)
        loss = criterion(out.squeeze(), data.y.squeeze())
        
        loss.backward()
        loss_all += loss.item() * data.num_nodes
        num_nodes += data.num_nodes
        optimizer.step()
    
    
    return loss_all / num_nodes

def test(loader):
      model.eval()
      test_correct = 0
      pred_len = 0
      for data in loader:
          out = model(data.x, data.edge_index, data.edge_attr)
          pred = out.argmax(dim= -1)  # Use the class with highest probability.
          test_correct += int((pred == data.y).sum()) # Check against ground-truth labels
          pred_len += pred.numel()
      return test_correct/pred_len

for epoch in range(1, 500):
    loss = train(epoch)
    tes_acc = test(test_loader)
    print(f'Epoch: {epoch:02d}, Loss: {loss:.4f}, '
          f'Test Accuracy: {tes_acc:.4f}')

I keep getting the following error:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-39-559f29094876> in <module>()
      1 for epoch in range(1, 500):
----> 2     loss = train(epoch)
      3     tes_acc = test(test_loader)
      4     print(f'Epoch: {epoch:02d}, Loss: {loss:.4f}, '
      5           f'Test Accuracy: {tes_acc:.4f}')

4 frames
/usr/local/lib/python3.7/dist-packages/torch_geometric/nn/conv/gcn_conv.py in forward(self, x, edge_index, edge_weight)
    177                     edge_index = cache
    178 
--> 179         x = x @ self.weight
    180 
    181         # propagate_type: (x: Tensor, edge_weight: OptTensor)

RuntimeError: mat1 and mat2 shapes cannot be multiplied (1234x2 and 1x32)

Could someone tell mw what is going wrong? I dont think there is any example of PDN here.

For more information, each graph looks something like this:

#Graph 1:
ea = np.array([.5,	.6,	.6,	.75,	.88,	.95,	.23,	.45,	.76,	.77,	.8,	.9,	.67,	.9,	.41,	.33,	.44,	.55,	.6,	.7,	.8,	.9,	.4,	.3,	.2,	.44,	.55,	.65,	.56,	.7,	.3,	.21,	.12,	.31,	.41,	.5,	.61,	.45,	.81,	.99,	.67,	.35,	.66,	.76,	.91,	.13,	.14,	.5,	.73,	.45,	.45,	.78,	.8,	.9,	.43,	.54,	.65,	.56,	.76,	.32,	.23,	.43,	.34,	.98,	.78,	.69,	.43,	.21,	.34,	0.56])
el = np.array([0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0,	0
])
clVec = np.array([4.743599, 1.910000, 3.300000, 4.403215, 2.180000, 4.193649, 4.742125, 3.485601, 3.523165, 3.099377, 3.323229, 3.320000, 5.895374, 4.603439, 3.876340, 2.581587, 3.357021, 3.299415, 2.640000, 6.490000, 3.409320, 3.540106, 3.639342, 3.066470, 3.394267, 2.769377,
                 1.880000, 2.793345, 5.831788, 4.473127, 3.173996, 2.859992, 2.639665, 3.600537, 3.111637, 4.153215, 2.070000, 3.693649, 1.260000,  2.300000, 3.593165, 1.772662, 1.450000, 3.185252, 5.009614, 3.330000, 1.240000, 2.173000, 3.471668, 3.170547, 2.927985, 1.840000,
                2.250000, 2.810000, 1.970000, 3.626847, 3.413165, 2.340000, 3.210431, 1.540000, 3.400000, 2.210000, 0.990000, 1.560000, 1.530000, 2.730000, 2.420000, 3.030000, 2.949320, 3.895488])
f_final = np.stack((ea,el), axis=1)
x = torch.from_numpy(f_final)
x = x.type(torch.FloatTensor)
y = torch.LongTensor([1 ,0, 0, 1, 0, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0,  1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1])
#y =  torch.reshape(def3.type(torch.LongTensor),(20,))
#y =  torch.reshape(y.type(torch.LongTensor),(20,1))
edge_index = torch.tensor([[0,2],[0,5],[0,10],[0,15],[0,23],[0,28],[0,38],[0,45],[0,53],[0,61],
                           [1,1],[1,3],[1,12],[1,19],[1,31],[1,43],[1,48],[1,57],[1,69],
                           [2,7],[2,17],[2,25],[2,33],[2,40],[2,47],[2,55],[2,59],[2,65],
                           [3,1],[3,4],[3,9],[3,13],[3,18],[3,21],[3,27],[3,30],[3,36],[3,42],[3,45],[3,51],[3,62],[3,67],
                           [4,5],[4,11],[4,14],[4,35],[4,44],[4,50],[4,64],
                           [5,6],[5,8],[5,16],[5,20],[5,26],[5,32],[5,41],[5,49],[5,54],[5,60],[5,66],[5,68], #12
                           [6,0],[6,10],[6,19],[6,22],[6,24],[6,29],[6,34],[6,37],[6,46],[6,52],[6,58],[6,63], #12
                           [7,0],[7,8],[7,12],[7,23],[7,39],[7,48],[7,56], #7
                           [8,1],[8,15],[8,25],[8,28],[8,31],[8,38],[8,43],[8,53],[8,61], #9
                           [9,0],[9,2],[9,7],[9,13],[9,21],[9,27],[9,40],[9,45], #8
                           [10,3],[10,11],[10,18],[10,30],[10,42],[10,47],[10,55],[10,57],[10,62],[10,68], #10
                           [11,4],[11,16],[11,26],[11,33],[11,50],[11,59],[11,65],[11,67],[11,69], #9
                           [12,0],[12,5],[12,10],[12,14],[12,20],[12,24],[12,34],[12,44],[12,49],[12,51],[12,54],[12,58],[12,64], #13
                           [13,6],[13,19],[13,22],[13,32],[13,36],[13,39],[13,46],[13,52],[13,56],[13,60], #10
                           [14,1],[14,7],[14,13],[14,25],[14,29],[14,35],[14,41],[14,44],[14,53],[14,63],[14,66], #11
                           [15,2],[15,8],[15,17],[15,23],[15,28],[15,37],[15,48],[15,61], #8
                           [16,3],[16,9],[16,12],[16,21],[16,26],[16,31],[16,38],[16,40],[16,43],[16,68], #10
                           [17,0],[17,3],[17,19],[17,31],[17,43],[17,48],[17,57],[17,69], #8
                           [18,7],[18,17],[18,25],[18,33],[18,40],[18,47],[18,59],[18,65], #8
                           [19,1],[19,4],[19,9],[19,13],[19,18],[19,21],[19,27],[19,30],[19,36],[19,42],[19,45],[19,51],[19,55],[19,67], #14
                           [20,3],[20,11],[20,14],[20,35],[20,44],[20,50],[20,64], #7
                           [21,8],[21,16],[21,20],[21,26],[21,32],[21,41],[21,49],[21,54],[21,60],[21,66],[21,68], #11
                           [22,0],[22,10],[22,19],[22,24],[22,29],[22,34],[22,37],[22,46],[22,52],[22,58],[22,63], #11
                           [23,0],[23,8],[23,12],[23,39],[23,48],[23,56], #6
                           [24,1],[24,6],[24,15],[24,25],[24,28],[24,31],[24,38],[24,43],[24,56],[24,61], #10
                           [25,0],[25,2],[25,7],[25,13],[25,17],[25,21],[25,27],[25,40],[25,45], #9
                           [26,3],[26,11],[26,18],[26,30],[26,42],[26,47],[26,55],[26,57],[26,68], #9
                           [27,4],[27,16],[27,26],[27,33],[27,59],[27,65],[27,67],[27,69], #8
                           [28,0],[28,5],[28,10],[28,14],[28,20],[28,24],[28,34],[28,44],[28,49],[28,51],[28,54],[28,58],[28,64], #13
                           [29,6],[29,19],[29,22],[29,27],[29,32],[29,36],[29,39],[29,46],[29,52],[29,56],[29,60], #11
                           [30,1],[30,7],[30,13],[30,25],[30,29],[30,41],[30,44],[30,53],[30,63],[30,66], #10
                           [31,2],[31,8],[31,17],[31,23],[31,28],[31,35],[31,37],[31,48],[31,61], #9
                           [32,3],[32,9],[32,12],[32,21],[32,26],[32,31],[32,38],[32,40],[32,43],[32,68], #10
                           [33,0],[33,3],[33,12],[33,19],[33,31],[33,43],[33,48],[33,57],[33,69], #9
                           [34,7],[34,17],[34,25],[34,33],[34,40],[34,47],[34,55],[34,59],[34,65], #9
                           [35,1],[35,4],[35,9],[35,13],[35,18],[35,21],[35,27],[35,30],[35,36],[35,42],[35,45],[35,51],[35,62],[35,67], #14
                           [36,5],[36,11],[36,14],[36,35],[36,44],[36,50],[36,64], #7
                           [37,6],[37,8],[37,16],[37,20],[37,26],[37,32],[37,41],[37,49],[37,54],[37,60],[37,66],[37,68], #12
                           [38,0],[38,10],[38,19],[38,22],[38,24],[38,34],[38,46],[38,58],[38,63], #9
                           [39,0],[38,8],[39,12],[39,23],[39,26],[39,29],[39,48],[39,58], #8
                           [40,1],[40,15],[40,25],[40,28],[40,31],[40,38],[40,43],[40,52],[40,53],[40,61], #10
                           [41,0],[41,2],[41,7],[41,13],[41,17],[41,27],[41,40],[41,45], #8
                           [42,3],[42,11],[42,18],[42,23],[42,30],[42,47],[42,55],[42,57],[42,68], #9
                           [43,4],[43,16],[43,26],[43,33],[43,50],[43,59],[43,65],[43,67],[43,69], #9
                           [44,0],[44,5],[44,10],[44,14],[44,20],[44,34],[44,49],[44,51],[44,54],[44,64],
                           [45,6],[45,19],[45,22],[45,30],[45,32],[45,36],[45,39],[45,46],[45,52],[45,56],[45,60],
                           [46,1],[46,7],[46,13],[46,25],[46,35],[46,44],[46,53],[46,58],[46,63],[46,66],
                           [47,2],[47,8],[47,45],[47,23],[47,28],[47,37],[47,48],[47,61],
                           [48,3],[48,9],[48,12],[48,21],[48,26],[48,31],[48,38],[48,40],[48,43],[48,68],
                           [49,0],[49,3],[49,12],[49,19],[49,31],[49,48],[49,57],[49,69],
                           [50,7],[50,17],[50,25],[50,33],[50,47],[50,55],[50,59],[50,65],
                           [51,1],[51,4],[51,9],[51,13],[51,18],[51,21],[51,27],[51,30],[51,36],[51,40],[51,42],[51,45],[51,62],[51,67],
                           [52,5],[52,11],[52,14],[52,35],[52,44],[52,50],[52,64],
                           [53,6],[53,8],[53,16],[53,20],[53,26],[53,32],[53,41],[53,49],[53,54],[53,60],[53,66],[53,68],
                           [54,0],[54,10],[54,19],[54,22],[54,24],[54,29],[54,34],[54,37],[54,46],[54,52],[54,58],[54,63],
                           [55,0],[55,8],[55,12],[55,23],[55,39],[55,48],[55,56],
                           [56,1],[56,15],[56,25],[56,28],[56,31],[56,38],[56,43],[56,53],[56,61],
                           [57,0],[57,2],[57,7],[57,13],[57,17],[57,21],[57,27],[57,39],[57,45],
                           [58,3],[58,11],[58,18],[58,30],[58,42],[58,47],[58,55],[58,57],[58,62],[58,68],
                           [59,4],[59,16],[59,26],[59,33],[59,50],[59,65],[59,67],[59,69],
                           [60,0],[60,5],[60,10],[60,14],[60,20],[60,24],[60,34],[60,44],[60,49],[60,51],[60,54],[60,58],[60,64],
                           [61,6],[61,19],[61,22],[61,32],[61,36],[61,39],[61,46],[61,52],[61,56],[61,60],
                           [62,1],[62,8],[62,13],[62,25],[62,29],[62,35],[62,41],[62,44],[62,53],[62,63],[62,66],[62,67],
                           [63,2],[63,8],[63,17],[63,23],[63,28],[63,37],[63,48],[63,61],
                           [64,3],[64,21],[64,31],[64,38],[64,40],[64,43],[64,68],
                           [65,0],[65,3],[65,9],[65,12],[65,19],[65,31],[65,43],[65,48],[65,57],[65,69],
                           [66,7],[66,17],[66,25],[66,33],[66,40],[66,47],[66,55],[66,59],[66,65],
                           [67,1],[67,4],[67,9],[67,13],[67,18],[67,21],[67,27],[67,30],[67,36],[67,42],[67,45],[67,51],[67,62],
                           [68,5],[68,11],[68,14],[68,35],[68,44],[68,50],[68,64],
                           [69,6],[69,8],[69,16],[69,20],[69,26],[69,28],[69,32],[69,41],[69,49],[69,54],[69,60],[69,66],[69,68]], dtype=torch.long)
edge_attr = torch.tensor([[.5],[.8],[.45],[.67],[.44],[.88],[.22],[.11],[.33],[.45],
                          [.1],[.4],[.75],[.65],[.65],[.77],[.8],[.22],[.78],
                          [.71],[.88],[.99],[.45],[.22],[.33],[.34],[.11],[.45],
                          [.11],[.34],[.35],[.22],[.44],[.76],[.56],[.44],[.23],[.11],[.34],[.21],[.11],[.3],
                          [.3],[.4],[1.1],[.61],[.75],[.65],[.13],
                          [.77],[.25],[.87],[.89],[.12],[.15],[.45],[.35],[.21],[.32],[.22],[.6], #12
                          [.3],[.35],[.44],[.75],[.81],[.91],[.7],[.8],[.2],[.15],[.18],[.28], #12
                          [.4],[.91],[.85],[.33],[.34],[.95],[.85], #7
                          [.22],[.67],[.5],[.8],[.2],[.11],[.33],[.4],[.33], #9
                          [.11],[.1],[.95],[.23],[.77],[.45],[.41],[.22], #8
                          [.5],[.33],[.11],[.55],[.3],[.45],[.32],[.31],[.11],[.7], #10
                          [.12],[.5],[.11,],[.66],[.44],[.33],[.12],[.33],[.77], #9
                          [.55],[.75],[.65],[.34],[.88],[.8],[.9],[.8],[.67],[.25],[.23],[.75],[.22], #13
                          [.66],[.43],[.71],[.54],[.23],[.32],[.11],[.41],[.33],[.45], #10
                          [.12],[.14],[.81],[.4],[.6],[.87],[.75],[.85],[.22],[.11],[.33], #11
                          [.4],[.1],[.12],[.45],[.7],[.5],[.76],[.11], #8
                          [.33],[.44],[.88],[.66],[.22],[.3],[.15],[.45],[.35],[.65], #10
                          [.1],[.4],[.65],[.65],[.77],[.8],[.22],[.78], #8
                          [.71],[.88],[.99],[.45],[.22],[.33],[.11],[.45], #8
                          [.11],[.34],[.35],[.22],[.44],[.76],[.65],[.44],[.23],[.11],[.34],[.21],[.56],[.11,], #14
                          [.3],[.4],[1.1],[.61],[.75],[.65],[.13], #7
                          [.25],[.87],[.89],[.12],[.15],[.45],[.35],[.21],[.32],[.22],[.6], #11
                          [.3],[.43],[.44],[.81],[.11],[.7],[.8],[.2],[.15],[.18],[.28], #11
                          [.4],[.91],[.85],[.34],[.95],[.85], #6
                          [.22],[.55],[.67],[.5],[.8],[.2],[.11],[.33],[.4],[.33],#10
                          [.11],[.3],[.1],[.95],[.23],[.77],[.45],[.41],[.22], #9
                          [.33],[.11],[.45],[.3],[.45],[.32],[.31],[.11],[.7], #9
                          [.12],[.5],[.11],[.56],[.33],[.12],[.33],[.77], #8
                          [.55],[.75],[.65],[.34],[.99],[.8],[.9],[.8],[.67],[.25],[.23],[.75],[.22], #13
                          [.66],[.43],[.71],[.12],[.54],[.23],[.32],[.11],[.41],[.33],[.45], #11
                          [.12],[.14],[.81],[.4],[.6],[.75],[.85],[.22],[.11],[.33], #10
                          [.4],[.1],[.15],[.55],[.7],[.44],[.5],[.76],[.11], #9
                          [.33],[.04],[.88],[.66],[.2],[.3],[.15],[.45],[.35],[.65], #10
                          [.1],[.4],[.76],[.65],[.65],[.77],[.8],[.22],[.78], #9
                          [.71],[.88],[.99],[.45],[.22],[.33],[.34],[.11],[.45], #9
                           [.11],[.34],[.35],[.22],[.44],[.76],[.56],[.44],[.23],[.11],[.34],[.21],[.11],[.3], #14
                           [.3],[.4],[1.1],[.61],[.75],[.65],[.13], #7
                          [.77],[.25],[.87],[.89],[.12],[.15],[.45],[.35],[.21],[.32],[.22],[.6], #12
                           [.3],[.35],[.44],[.75],[.81],[.7],[.2],[.18],[.28], #9
                           [.4],[.91],[.08],[.33],[.18],[.09],[.95],[.85], #8
                           [.22],[.67],[.5],[.8],[.2],[.11],[.33],[.16],[.4],[.33], #10
                           [.11],[.3],[.1],[.05],[.23],[.45],[.41],[.22], #8
                           [.5],[.33],[.11],[.43],[.05],[.45],[.32],[.31],[.7], #9
                           [.12],[.5],[.11],[.56],[.44],[.33],[.12],[.33],[.77],
                           [.55],[.75],[.65],[.34],[.88],[.9],[.67],[.25],[.23],[.22],
                          [.66],[.43],[.71],[.66],[.54],[.23],[.32],[.11],[.41],[.33],[.45],
                          [.12],[.14],[.81],[.4],[.87],[.85],[.22],[.44],[.11],[.33],
                          [.4],[.1],[.12],[.45],[.07],[.5],[.76],[.11],
                          [.33],[.44],[.88],[.06],[.22],[.3],[.15],[.45],[.35],[.65],
                          [.1],[.4],[.75],[.51],[.65],[.8],[.22],[.78],
                          [.71],[.78],[.99],[.45],[.33],[.34],[.11],[.45],
                          [.11],[.34],[.35],[.22],[.44],[.76],[.56],[.44],[.23],[.44],[.11],[.98],[.11],[.3],
                          [.3],[.4],[1.1],[.61],[.75],[.65],[.13],
                          [.55],[.25],[.87],[.89],[.12],[.15],[.45],[.35],[.21],[.32],[.22],[.6],
                          [.3],[.35],[.44],[.75],[.81],[.99],[.7],[.8],[.2],[.15],[.18],[.28],
                          [.4],[.99],[.85],[.33],[.34],[.95],[.85],
                          [.22],[.67],[.5],[.8],[.2],[.11],[.33],[.4],[.33],
                          [.11],[.3],[.1],[.95],[.33],[.77],[.45],[.41],[.22],
                          [.5],[.33],[.11],[.45],[.3],[.45],[.32],[.31],[.11],[.7],
                          [.12],[.5],[.11],[.56],[.44],[.12],[.43],[.77],
                          [.55],[.75],[.65],[.34],[.88],[.8],[.9],[.8],[.67],[.25],[.23],[.75],[.22],
                          [.66],[.43],[.71],[.54],[.23],[.32],[.11],[.41],[.33],[.45],
                          [.12],[.13],[.81],[.4],[.6],[.87],[.75],[.85],[.22],[.11],[.33],[.41],
                          [.4],[.1],[.12],[.45],[.7],[.5],[.76],[.11],
                          [.33],[.66],[.3],[.15],[.45],[.35],[.65],
                          [.1],[.4],[.45],[.75],[.55],[.65],[.77],[.8],[.22],[.78],
                          [.71],[.88],[.99],[.45],[.22],[.33],[.34],[.11],[.45],
                          [.11],[.34],[.35],[.22],[.44],[.76],[.56],[.44],[.23],[.11],[.34],[.21],[.11],
                          [.3],[.4],[1.1],[.61],[.75],[.65],[.13],
                          [.77],[.25],[.87],[.89],[.12],[.1],[.15],[.45],[.35],[.21],[.32],[.22],[.6]],  dtype=torch.float)
data = Data(x = x, edge_index = edge_index.t().contiguous(), y = y, edge_attr=edge_attr)
data_list.append(data)

[rusty1s]

There is already a PDNConv layer in PyG master which you can also make use of :)

Your issue seems to be related to a feature mismatch:

self.convolution_1 = GCNConv(data.x, node_filters)

Note that the GCNConv needs to be initialized with data.x.size(-1) instead of data.x.

[rusty1s]

Are your issues resolved? It seems like you deleted your post.