Set_rng_seed in pyro


I am wondering how pyro.set_rng_seed alters the parameters of the guide distribution?
If I change the pyro seed with pyro.set_rng_seed() right before Predictive(my_pyro_model, guide=my_trained_guide), would it somehow alter the trained parameters in my_trained_guide, thus preventing proper predictive samples to be generated from my_trained_guide?

Thank you,

Interesting! I believe pyro.set_rng_seed only changes how random samples are generated. Probably your parameters are still the same, but guide generated different latent values when you changed the rng seed? Could you check it again?

So I am working on a Pyro neural network model, and the validation error from the training phase is very small, indicating the guide is very well trained. But I am getting 0% accuracy rate when I try to predict the y’s using the trained guide. The only difference in the way my Pyro model is implemented during the training phase vs. Prediction phase is that in the prediction phase, I change the pyro random seed with set_rng_seed() before I execute the lines Predictive(...) and predictive_obj.forward(...). So I am thinking the reason why I am getting 0% accuracy rate for my prediction is due to the fact that the parameters of my well-trained guide is altered by the change in the random seed…

I know that I am likely to be wrong, but this is the only explanation that I can come up with for the 0% accuracy rate in prediction…

I don’t know why this is happening…:frowning: