MultiCoreBPRMF different results

[jordansilva]

The MultiCoreBPRMF return different results training with the same parameters and fixing the Random Seed. Still, BPRMF (without MultiCore) gets the same results when the Random Seed is fixed.

[jordansilva]

Parameters
MultiCoreBPRMF num_factors=10 bias_reg=0 reg_u=0.06 reg_i=0.06 reg_j=0.006 num_iter=25 learn_rate=0.05 uniform_user_sampling=False with_replacement=False update_j=True max_threads=100

[zenogantner]

That looks like a race condition. Not sure whether we will fix this. Should we document the behavior?