Infinite loop when training on O2 with multiplicity 3

[arnon-1]

Not sure if this is expected behaviour, but the default implementation for the initial electron configuration doesn't seem to work in all cases.
The script tries to assign a number of electrons to each atom according to the charge and ground state spin of the individual atoms (as mentioned in the ferminet paper) but this fails in some cases.
The main issue is that this behaviour is hardcoded and doesn't seem to be easily changeable without modifying the source code.
Example to reproduce:

import sys

from absl import logging
from ferminet.utils import system
from ferminet import base_config
from ferminet import train

# Optional, for also printing training progress to STDOUT.
# If running a script, you can also just use the --alsologtostderr flag.
logging.get_absl_handler().python_handler.stream = sys.stdout
logging.set_verbosity(logging.INFO)

# Define O2 molecule
cfg = base_config.default()
cfg.system.electrons = (9,7)  # (alpha electrons, beta electrons)
cfg.system.molecule = [system.Atom('O', (0, 0, -1)), system.Atom('O', (0, 0, 1))]

train.train(cfg)

[jsspencer]

Sorry, this used to be ok. I'll add a fallback strategy -- don't have time for a complete fix, as selecting particle initialisation is hard -- though increasing the number of pretraining and MCMC burn-in iterations will probably be a good idea for such cases (or at least carefully monitoring MCMC and convergence).

[arnon-1]

Ok, thanks