Different results on the same data with the same seed

Modeling
1

Hello everyone!

I am using a model based on STAN, and found out that the results are different with the same data, even if the same seed is used. It also seems that the amount of different results is not endless, but just two or three. What can it be? I thought maybe the random walk comes to different local minimums, but could it be the case with the same seed?

The model params are these:

    STAN_MODEL_PARAMETERS: Tuple[Tuple[str, Any]] = (
        ("iter_sampling", 1000),
        ("chains", 2),
        ("iter_warmup", 1000),
        ("thin", 1),
        ("seed", 1),
        ("adapt_delta", 0.95),
    )
2

I once had a problem like this, and the initial values were to blame. I am not sure whether the seed also covers the initial values, but you might define the initial values yourself to make sure they are always the same.

1 Like