Bullet-proof way to `update_log`

As of right now the MLE_Logger module is updated based on two lists of time measurements and statistic measurements:

logger.update_log(clock_tick, stats_tick, model=None,  plot_to_tboard=None,  save=False)

This is not bullet proof and can be hard to debug. I would like to have this be dictionary based. At the start of each update update_log should check if all keys are there. Afterwards, the data is added to the .hdf5 log. Thereby, the order of how the data is given as an input does no longer matter.

Thoughts:

If some key is not there, fill in with NaN or some missing value? How does this affect meaning of logs in gridsearch?

mle-infrastructure / mle-toolbox

Bullet-proof way to `update_log` #50