4 ms·
The problem with frameworks like this is when you try to do something which isn't strictly standard, or lack understanding on whatever the framework is actually
by baalimago 7y ago
The problem with frameworks like this is when you try to do something which isn't strictly standard, or lack understanding on whatever the framework is actually doing. Things will no doubt break, and then you have a 15 layered framework stack trace to try to debug.
You can't dodge knowledge with frameworks, only use frameworks to effectivize your knowledge.
- seanmcdirmid 7y agoThis is exactly the premise of the paper, which isn’t presenting a framework, rather a language for general open ended computation for ML.
- ChrisRackauckas 7y agoI haven't had that problem with Julia-based probabilistic programming frameworks. I stick functions from pre-existing packages into Turing.jl and it seems to work fine as long as the library is compatible with AD.