Building cov1d model with residual block + hypermeter tunner

2

This definition:

        self.blocks = nn.Sequential(
            *[ResidualBlock(in_channels, out_channels, kernel_size, l2_lambda=l2_lambda) for _ in range(num_blocks)]
        )

won’t work for multiple ResidualBlocks since each of these will expect an input with in_channels channels while the previous one will return an activation with out_channels.
You could fix it e.g. via:

        self.blocks = nn.Sequential(
            ResidualBlock(in_channels, out_channels, kernel_size, l2_lambda=l2_lambda),
            *[ResidualBlock(out_channels, out_channels, kernel_size, l2_lambda=l2_lambda) for _ in range(num_blocks-1)]
        )

but would need to double check if the “internal” ResidualBlocks should use out_channels for their input and output.

1 Like