Flat Distribution

Hi pymc community,

I hope this question isn’t too simple to ask on this board. I bumped into pm.Flat('foo') while learning PyMC3, and have no clue what it is used for. I have read the documentation to no avail. Googling “Flat Distribution” does not seem to return anything relevant either. Besides

Uninformative log-likelihood that returns 0 regardless of the passed value.

, is there a more intuitive explanation for what this function/distribution is used for?

Much thanks from a beginner!

1 Like

For random variable that defined using a Flat distribution, you can understand it as a free parameter that follow a uniform distribution on (-inf, inf). It’s also called an uninformative prior (some what misleading as Flat prior is also informative under some transformation).
In most case we dont recommend using it as using a Normal distribution with large sigma (i.e., a weakly informative prior) is usually much better re model convergence.


free parameter that follow a uniform distribution on (-inf, inf)

That makes sense, 0 everywhere is essentially a uniform on R, and hence flat! Thanks for the quick, concise response junpenglao.

1 Like