Set random number seed for generator

[koheiw]

To replicate the results, it is not enough to initialize the output matrix with the same random number. I made changes my experimental package to make analysis replicable when threatds = 1.

koheiw/wordvector@8dc3d78

[jwijffels]

Thanks for the heads up.
When I did some tests, the training run was completely reproducible with threads = 1 but not with threads > 1.
Is the training run also completely reproducible with threads > 1 in your package?

[koheiw]

If I set random seed, the result looks very similar when threads > 1, but not identical. I don't think it is possible to make results identical in parallel computing.