Can weights from leela-zero training be used to bootstrap the ScalphaGoZero network?

[barrybecker4]

Question

I assume that ScalphaGoZero will not play very well without a whole lot of training first. Leela-zero has been training its network for over a year now. Both of these projects claim to use the same residual convolutional neural net architecture as the original apha-go-zero paper. Would it be possible to just use the weights already learned from the many months of leela-zero crowd-sourcing as a starting point for ScalphaGoZero learning? If so, what are the steps to do it?

[maxpumperla]

@barrybecker4 we "just" need to understand how the weights are stored. If you manage to read them from Python and store them, for instance, in npy format, we can definitely load them as Nd4j arrays and eventually set them as weights for a corresponding ScalphaGoZero network.

[maxpumperla]

Let me know if you want to tackle this. I can point you to the right places, but don't currently have time to get my hands dirty, I'm afraid. :/

[barrybecker4]

Thanks. I'm not quite ready to look at it at the moment but hope to get to it eventually.