'max_norm' in train interface

[lyangfan]

I have encountered the problem of gradient explosion in the training process. In order to solve this problem, could you add the 'max_norm' parameter and corresponding functions into the training interface?

Thanks a lot

[clementchadebec]

Hi @lyangfan ,

Thank you for this request. This would indeed be a great improvement to the lib! I will try to add this feature soon. In any case, this library is open to contributors so feel free to make a proposal and send a Pull Request if you want to contribute. I would be very happy to review it :)

Best,

Clément