Can't generate with biases about tensorflow-wavenet

[ibab]

There's a bug in how the arguments are handled. I've fixed this as part of #88.

[jyegerlehner]

OK thx ibab. I'll check that #88 fixes it, and if so close the issue.

[ibab]

Adding biases to the fast generation is also quite easy. I could do that next.

[jyegerlehner]

Is it? The thought crossed my mind when I saw it there and was putting in Xavier init, but I didn't understand all that black magic so didn't want to touch it. Better you than me. :)

[ibab]

I've added biases (and tests) to the fast generation in #88.

[jyegerlehner]

@ibab Is #88 meant to be backward compatible with existing checkpoint files? I did a naive merge and tried to generate, and got errors about tensor size mismatch. First dimension 1 vs. 2. Batch size change?

I'm guessing to test this branch I need to restart training from scratch and use that model to generate from.

[ibab]

@jyegerlehner: I've just realized that this was an indication of the bug that I needed to fix :D
Generating with biases should be working now, so I'm closing the issue.