Meta-RL icon indicating copy to clipboard operation
Meta-RL copied to clipboard

Published 20 hours ago • verified publisher icon awjuliani

'NoneType' object has no attribute 'model_checkpoint_path'

Open sunchipsster1 opened this issue 5 years ago • 7 comments

Hello and happy 2020! I am trying to run A3C-Meta-Bandit and run into the error below. All the cells are running fine, except for the final cell of the python notebook script which produces this error. Is there some folder that is missing? Thank you!

image

sunchipsster1 avatar Jan 17 '20 20:01 sunchipsster1

Hi @sunchipsster1

What version of tensorflow are you using? This notebook were originally created with a much older version, and as such this code may not work in more recent versions.

awjuliani avatar Jan 21 '20 18:01 awjuliani

Hi Arthur! Thank you so much for your code which is so beautiful and easy to understand! Your blogs have been an amazing resource for the community. I am using Python3.7.3. Do you have an idea, conceptually speaking, what may cause this error and here the solution might lie?

sunchipsster1 avatar Jan 21 '20 20:01 sunchipsster1

Hi @sunchipsster1

There shouldn't be a problem with that python version, but can you share the TensorFlow version?

awjuliani avatar Jan 21 '20 21:01 awjuliani

Yes, the Tensorflow version is 1.14.0. Thank you!

sunchipsster1 avatar Jan 21 '20 21:01 sunchipsster1

I would recommend trying an earlier version of 1.7 or earlier.

awjuliani avatar Jan 22 '20 00:01 awjuliani

Hi Arthur, thank you for your help and apologies for the continued questions! I:

  1. tried installing an earlier version of Tensorflow, but versions < 1.13.1 no longer seem to be available. I tried with 1.13.1, but this still does not work.
  2. There are links to earlier versions on other websites. I tried these, but they did not work properly.
  3. I have previously run your Deep RL codes (this is how I learned how to do Deep RL :) ... these worked fine, and have the same code using saver, model_checkpoint_path, etc. so I suspect that 1.13.1 should work fine with your meta-RL code here too.

Therefore, I resolved to diagnose what could be slightly amended in your code to make it work in 1.13.1. I have attached my current settings (attached image) which -- unfortunately do not produce training or save data. One thing I noticed which might help solve this mystery is the following: ... In your blog post, you said that you ran the Bandits problem on 100 trials of 20 000 episodes. Where is the setting for this, in your code? There does not seem to be any code to specify the number of training episodes...

Thank you!

image

sunchipsster1 avatar Jan 22 '20 23:01 sunchipsster1

Hi @sunchipsster1,

It has been a while since I have looked at the code, but I believe the workers are set to run indefinitely, and only terminate when the process has been interrupted by the user.

awjuliani avatar Jan 23 '20 19:01 awjuliani