fani-lab / SEERa

A framework to predict the future user communities in a text streaming social network based on the users’ topics of interest.
Other
4 stars 5 forks source link

DynamicGEM issues #38

Closed soroush-ziaeinejad closed 2 years ago

soroush-ziaeinejad commented 2 years ago
  1. When calling AERNN, it calls future library which is available from python 3.7. Thus, in yml file, python 3.7 or 3.8 should be mentioned.
soroush-ziaeinejad commented 2 years ago

tensorflow-gpu==1.11.0 is not available for python 3.7 and newer!!

soroush-ziaeinejad commented 2 years ago

@hosseinfani

Can you please run: from dynamicgem.embedding.dynAERNN import DynAERNN on the system that you installed dynamicgem and let me know the result?

Thanks

soroush-ziaeinejad commented 2 years ago

I think commenting "from future import annotations" out in "site-packages\seaborn-0.12.0b2-py3.6.egg\seaborn\axisgrid.py" will resolve the problem

soroush-ziaeinejad commented 2 years ago

Now it's training

hosseinfani commented 2 years ago
  1. When calling AERNN, it calls future library which is available from python 3.7. Thus, in yml file, python 3.7 or 3.8 should be mentioned.

but the dynamicgem works in python 3.6. How this is possible?

hosseinfani commented 2 years ago

tensorflow-gpu==1.11.0 is not available for python 3.7 and newer!!

I know. we deal with gpu issues in the next round of code refactoring.

soroush-ziaeinejad commented 2 years ago
  1. When calling AERNN, it calls future library which is available from python 3.7. Thus, in yml file, python 3.7 or 3.8 should be mentioned.

but the dynamicgem works in python 3.6. How this is possible?

The AERNN calls another library (seaborn) which import future library and this library is not available for python 3.6 and older. I commented out import future and it worked fine. Now the cpl part 2 (the bottleneck) is running on the workstation.

soroush-ziaeinejad commented 2 years ago

@hosseinfani Also the apl is almost ready but it needs user clusters. After running of cpl and saving user clusters, I can continue work on apl.

hosseinfani commented 2 years ago

from dynamicgem.embedding.dynAERNN import DynAERNN

@soroush-ziaeinejad I found the reason. DynamicGem wants seaborn>=0.9.0. The current version of seaborn (0.12) needs python 3.7. I believe if you put seaborn==0.9.0 or 0.11.0, in our environment.yml and requirement.txt, dynamicgem sees that and won't look for latest version. I checked it and it worked.

soroush-ziaeinejad commented 2 years ago

Graph embedding results

Setting 1:

Epochs: 150 Dimension: 128 embedding.max(): 0 embedding.min(): 0

Setting 2:

Epochs: 10 Dimension: 64 embedding.max(): 0.0017 embedding.min(): -0.0019

Setting 3:

Epochs: 10 Dimension: 128 embedding.max(): 0.0054 embedding.min(): -0.0061