Loading ASE calculator from model trained with basic config template fails about apax HOT 5 CLOSED

Hi Ali,

Sorry for the late response, we were all attending a conference these last few days.

The Path you need to specify is the one to ".../directory/experiment".
We don't assume a particular input file name and directory may contain multiple experiments (different model trainings).
Hence, pointing the ASE calculator to the directory with the configuration file is not sufficient.

We could however provide a better error message.

Best
Moritz

from apax.

No worries, Moritz, I hope the conference went well.

I have tried that directory and get the same error, unfortunately...
From my config file, I should be pointing the ASECalculator class to models/apax, (from the directory and experiments entries in the yaml file in the data heading), but doing so results in the same error mentioned above. This directory has a config.yaml file generated by apax with more details than my initial input file.

Any idea what could be going wrong?

from apax.

Currently, model loading assumes that you are in the same directory the training was started from.
so if you are in mydir and run a training from there with

directory: models
experiment: exp

then you should be able to to load a model into ASE via
ASECalculator("models/exp") from a python file located in mydir.

I am not sure if you are attempting to load the model from a different relative path, but I suspect that might be the issue here.
It will be fixed by #212 .

The setup described above is what I use for working with apax standalone.
We usually hand over path handling to our workflow manager, IPSuite, so it's certainly true that training at and loading from arbitrary locations was not well tested until now.

from apax.

Ah, yes, I was trying to load it from one directory above the training directory. Other codes I use (and want to compare against) don't interface with IPSuite so I wasn't using it, hence this issue.

Thank you for your help!

from apax.

No problem! thanks for raising this issue.

from apax.