Closed ayushkarnawat closed 4 years ago
After a bit of testing to determine where the NaN values are being introduced into the network, it seems that the issue is that there are NaN
s in the preprocessed dataset. This means, as the data is getting processed by a forward pass through the network, the output result of the NaN
values cannot be properly computed (specifically by the ReLU activation function above). Hence, there is an error output in that portion of the code.
We still have to determine where exactly the NaN
values are being introduced in the preprocessed dataset (aka during when in the computation of the mol features are there NaN
values).
The issue is related to the torch.return_types.max
object. To obtain a Tensor
, simply call .values
on the object.
I am trying to run a forward pass through the EmbeddedGCN model using
Current behavior
Expected behavior
The model runs a forward pass and outputs a real value.