Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Can't load model when using HER + SAC

See original GitHub issue

When loading a model using HER + SAC I get the following error:

Traceback (most recent call last):
  File "main.py", line 35, in <module>
    model = HER.load('test.zip', env = env)
  File "/home/tirafesi/.local/lib/python3.6/site-packages/stable_baselines/her/her.py", line 148, in load
    data, _ = cls._load_from_file(load_path, custom_objects=custom_objects)
  File "/home/tirafesi/.local/lib/python3.6/site-packages/stable_baselines/common/base_class.py", line 652, in _load_from_file
    data = json_to_data(json_data, custom_objects=custom_objects)
  File "/home/tirafesi/.local/lib/python3.6/site-packages/stable_baselines/common/save_util.py", line 120, in json_to_data
    base64.b64decode(serialization.encode())
ModuleNotFoundError: No module named 'numpy.random._pickle'

> pip freeze | grep -E 'numpy|stable'
numpy==1.17.3
stable-baselines==2.10.0a0

Issue Analytics

State:
Created 4 years ago
Comments:5

Top GitHub Comments

1reaction

siferaticommented, Mar 12, 2020

Yes, that was definitely it. Updating my local version of numpy to match the one on the remote computer fixed the issue.

0reactions

Miffylicommented, Apr 1, 2020

@erwincoumansepic

If you are not able to load the full model, you can still try load_parameters function to load the network weights of that saved model. You need to create model similar to one stored in the model file manually, though.

As for the support independent of numpy version: Saving still uses pickling for some parts of the models, which is often pretty sensitive to library versions. In v3, with PyTorch backend, we will have more robust saving/loading.

Top Results From Across the Web

Facts could not be loaded - error with SAC model

One of the users is getting an error 'Facts could not be loaded' while accessing a model based on an import data connection...

SAC — Stable Baselines 2.10.3a0 documentation

The SAC model does not support stable_baselines.common.policies because it uses double q-values and value estimation, as a result it must use its own...

Retraining model after loading not working while following #30

Check what is the performance of model upon saving with some evaluation games (no training), then load the policy and evaluate again.

Resolved - Can't load trained model for inference - Unity Forum

Please provide : The model you are trying to use. The versions of C# ml-agents you are using. A print out of the...

Use SAC to troubleshoot your Windows instance

Limitations. If you enable SAC on an instance, the EC2 services that rely on password retrieval will not work from the Amazon EC2...