Set random seed for approximate inference

[arose13]

I'm trying to set the random seed for approximate inference. It appears as if it is not possible but I'm not sure if this is true.

(The method being used is ADVI)

self._approx = pm.fit(n=100000, method=approx.lower(), random_seed=42)  # type: pm.Approximation
self.trace = self._approx.sample(draws=n_samples)  # HERE I CANNOT SET A SEED

I assume this is not impossible in principle. Let me know if there's anything I can do that can help

[ferrine]

This is not possible in the current API. We access theano symbolic random generator. It is possible to implement your case, however. It would require some additional theano manipulations in pymc3:)

[arose13]

@ferrine Okay understood.

It's not a high priority I so worse-case I'll see if I can make a pull request for this feature.