Filter dimension for the first conv1d operation

[wwang2]

Hi! Thanks for the implementation

I think there might be some problem when implementing the conv1d layer here:

molencoder/molencoder/models.py

Line 56 in 1d7e208

self.conv_1 = ConvSELU(i, 9, kernel_size=9)

I think the definition of o' and i' are swapped. The input filter dimension should the number of possible characters (35), and `c' should be the padded SMILES sequence length.

Just want to confirm this detail. I can submit a PR if that is correct.