Dedicated MD scheduled 18-20:00. Operational test on flattop.
Logbook entry: https://logbook.cern.ch/elogbook-server/GET/showEventInLogbook/4149736
Post-MD actions
- Extract BCTs and spill quality values from NXCALs to compare performances [priority:: high] [completion:: 2024-10-14]
Model finetuned on data from MD 2024-09-04, and then evaluated on the data from this MD yielded the following:
Model seems to have fitted perfectly to the cycle sequences seen in the MD previously, since the same cycles compositions were used.
Below is the same predictions but with a model trained on more varied data, amongst others with MD1 turned off.
This would suggest that our model learns the data, but not the hysteresis. I.e. We would need to train our model on specific operational cycle sequences, which are then used in testing.