Open jordansilva opened 7 years ago
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
That looks like a race condition. Not sure whether we will fix this. Should we document the behavior?
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.