-
-
Notifications
You must be signed in to change notification settings - Fork 229
Testimonials
Chris Fonnesbeck edited this page Apr 12, 2016
·
1 revision
I think many of pyMC’s advantages stem from the fact that it is written in a living language. For example, I was able to write myself the routines to read data directly from my database and use the data with pyMC. I’m sure there are people who could do the same with BUGS and component-pascal. However, the only reason I can think of to learn Component-Pascal is so that I can make better use of BUGS. When I learned PYTHON, many useful libraries were made available to me.
I have created my own functions and distributions with BUGS and there was a lot of difficult syntax to follow. These same tasks were very simple with pyMC.
With pyMC, I can break my model into components and reuse those components.
Since pyMC is a library instead of an entire package, it is easier to use it in conjunction with other software (file-reading, graphing, output, further data-analysis, etc).
I think anybody with a programming-background will find pyMC to be powerful, flexible and reasonably easy to learn. This type of person is likely to find WINBUGS constraining.
WINBUGS is a good tool for somebody who wants to use Markov Chains but is not comfortable in a programming environment. Given this constraint, the creaters of WINBUGS have done a remarkable job.
From a configuration-control point of view, I find it much easier to manage versions of models, versions of data and resulting outputs with pyMC.
Comparing PyMC to the popular WinBugs package, my top 3 are:
Having been likely the only person to have learned their Bayes solely with PyMC I have found that the absence of black-boxes (e.g. latents in WinBUGS) has forced me to know how each component works in any given model. I think this is invaluable when learning to understand and use Bayesian methods day to day.
This is very, very nice. I intend to use PyMC soon in my academic work, about approximate inference on continuous time stochastic processes with discrete time observations by using some statistical physics tools. Maybe it’s all there in the Gaussian Processes package, maybe I need to make some new stuff of my own. But in any case it would be very nice to share the code with my coworkers and have a solid package, easy to install and use, always updated and reliable such as PyMC. The main quality this package I think is the fact that, besides all the technical things and the open source license, we can count on the developers to be always at hand, solving problems and helping even with some basic questions most developers wouldn’t have patience to answer. Congratulations for the new version. I’m eager to use it and will use it soon.
Wayne Hajas
Comparing PyMC to the popular WinBugs package, my top 3 are:
- It is actually possible to debug PyMC (including interactively)
- PyMC allows real programming (e.g., variable reassignment in loops)
- Much easier access to relational databases
Mike Conroy
Having been likely the only person to have learned their Bayes solely with PyMC I have found that the absence of black-boxes (e.g. latents in WinBUGS) has forced me to know how each component works in any given model. I think this is invaluable when learning to understand and use Bayesian methods day to day.
Aaron MacNeil
This is very, very nice. I intend to use PyMC soon in my academic work, about approximate inference on continuous time stochastic processes with discrete time observations by using some statistical physics tools. Maybe it’s all there in the Gaussian Processes package, maybe I need to make some new stuff of my own. But in any case it would be very nice to share the code with my coworkers and have a solid package, easy to install and use, always updated and reliable such as PyMC. The main quality this package I think is the fact that, besides all the technical things and the open source license, we can count on the developers to be always at hand, solving problems and helping even with some basic questions most developers wouldn’t have patience to answer. Congratulations for the new version. I’m eager to use it and will use it soon.
Rafael Calsaverini