Experiments failed to be reproduced about continual-learning-baselines HOT 4 CLOSED

Ciao Andrea -

Thank you for putting this project together. I am trying to run and replicate these experiments but I am having issues running the scripts as described. It's failing to import experiments. Which root I should be in?
Also, I want to do the experiments with different Optimizers, can you please advise?

from continual-learning-baselines.

Hi @hsalkhatib for the moment few strategies have different results than expected. Your problem however is related to the PYTHONPATH. You need to add the project to your PYTHONPATH in order to be able to import it. Please, follow the instructions in the README and you should be fine. For example, to run EWC on Permuted MNIST you can simply launch

PYTHONPATH=${PYTHONPATH}:/path/to/continual-learning-baselines-project python experiments/permuted_mnist/ewc.py

If you want to change optimizer you can just change the ewc.py content (my suggestion is to fork/clone the repo and apply your modifications).

Hope this helps!

from continual-learning-baselines.

Thank you Andrea, I appreciate your quick response - Hassan

from continual-learning-baselines.

All the experiments are now running fine, except for COPE on Split MNIST. Issue is tracked on the main Avalanche repo ContinualAI/avalanche#1054

from continual-learning-baselines.