Bigger synthetic dataset

[tkukurin]

Nice work, thanks for publishing this reproducible version.

I was wondering whether:

1. Could you clarify how the train.bin data was seeded? How is it different to the data whose results are reported in the paper?
2. Is it possible to publish a bigger train.bin dataset?

Thanks!