Hi @Hrima89,
To be a bit more specific, I’d say it’s less PyMC’s implementation of the uniform (i.e pm.Uniform) than the uniform priors in general that are discouraged.
In short, they induce a nasty geometry to sample from, and we actually almost always have some information on our analysis. In addition, a flat prior doesn’t mean it’s non-informative.
In your case, if you’re trying to enforce a positivity constraint, you can look into Gamma distributions, or Exponential and HalfNormal (usually used for scale parameters). Here are recommendations for good prior choices.
Hope this helps 