Mimi Framework

Package incompatibility from Mimi.jl


I plan to use the mimi package but there seems to be incompatibility to another package I’m using. The installation of Mimi.jl/MimiFUND.jl will downgrade my installed latest version ReinforcementLearning.jl. Please check the error and reply from the administrator of ReinforcementLearning.jl below. In the last reply, he mentioned the dependence from MimiFUND.jl on Distributions and DataStructures is not latest.
JuliaReinforcementLearning/ReinforcementLearning.jl#244 (comment)

Could you please help me with this? Thank you!

Hi @lguo,

I’m so sorry you’re dealing with this, it looks like ReinforcementLearning.jl has very tight bounds on the newest releases of certain packages and thus you’re hitting an incompatibility with DataStructures.jl, I think Distributions.jl is ok for now but I’ll spare you the details.

I should be able to get this fixed up quickly and will ping you as soon as I do.


No worries. Thank you!

Btw, there seems to be more incompatibility issues when I install the latest machine learning package Flux.jl and GPU processor CUDA.jl. The later will downgrade my MimiFUND version. Below are the pictures.

Thank you very much!

I’ve talked with the downstream developers and fixed the DataStructures.jl problem related to Mimi, now I’ll keep working on this!

Just to be clear, the packages you want to work with in the same environment are

Mimi v1.2.2
MimiFUND v3.12.1
CUDA v3.1.0
Flux v0.12.3
ReinforcementLearning v0.8.0


Ok @lguo I have stared at a bunch of compatibility trees and requested a few tags by developers, normally it’s not this tricky but these packages just have a lot of dependencies, and really want the most recent releases of those, but that’s fine we can do it!

I have a shortcut for you if you want to use CUDA, Mimi, and MimiFUND immediately but if there’s not a huge rush you can avoid any trickiness of checking out branches and just wait a day for the tags to flow through and I’ll ping you.

If you’d rather work with MimiFUND v3.12.0, Mimi v1.1.2, and CUDA v3.1.0 in the meantime let me know.

Thank you! I can wait for a day. The only package I want to work with Mimi for now is the ReinforcementLearning@v0.8. Flux, CUDA and some other packages are used in the ReinforcementLearning package. Thus far I can find the incompatibility for the three.

Great, yes I figured out the two sticking points and once those are properly tagged then you should be all set.

Cool! Thank you! I update the packages in Julia and it works. I will wait for your update on the RL package.

Ok, I just opened a fresh environment and using

(temp) pkg> add Mimi, MimiFUND, ReinforcementLearning

and then get

(temp) pkg> st
      Status `~/temp/Project.toml`
  [e4e893b0] Mimi v1.2.2
  [b3ba11de] MimiFUND v3.12.1
  [158674fc] ReinforcementLearning v0.8.0

you should be set with those three together! Let me know how that goes

Yes, it works now. Thank you so much!

1 Like