Geometric temporal GConvLSTM

Hello everyone,

I’ve been working on a project involving graph convolutional LSTM models in PyTorch, and I’ve encountered a shape mismatch error that I’m having trouble resolving.

I have implemented a GConvLSTMModel using the source code for torch_geometric_temporal.nn.recurrent.gconv_lstm available at torch_geometric_temporal.nn.recurrent.gconv_lstm — PyTorch Geometric Temporal documentation. This is my model:

model_GconvLSTM= GConvLSTMModel(in_channels=6, out_channels=22, num_layers=6, K=2)

During training, I have the error:

/usr/local/lib/python3.10/dist-packages/torch_geometric/nn/dense/linear.py in forward(self, x) 130 x (torch.Tensor): The input features. 131 """ --> 132 return F.linear(x, self.weight, self.bias) 133 134 @torch.no_grad() RuntimeError: mat1 and mat2 shapes cannot be multiplied (22x6 and 22x22)

In the forward method of the model, the data tensor has dimensions (batch_size=1, num_nodes=22, num_features=6, num_time_steps=5). Here, num_nodes is the number of graph nodes, num_features represents the number of features for each node, and num_time_steps indicates the number of graph representations over time.

I’ve double-checked the shapes of my tensors and input data, and I’m fairly confident that the issue arises from a shape mismatch somewhere in the model architecture. I suspect that I might be passing incorrect dimensions or not handling the temporal dimension properly.

Could someone kindly guide me on how to properly configure the input data shape and architecture to resolve this error? I’d greatly appreciate any insights or suggestions to help me move forward