Some default | hard-coded values are not consistent between scripts/save_batches.py (making batches) and
run.py (setting model dimensions). In particular, running scripts/save_batches.py in non-production model sets the sat_delay to 30 minutes (hard-coded, which is reasonable). However, running thereafter run.py provides as default min_sat_delay_minutes: Optional[int] = 15, leading (I believe) to a mismatch in dimensions expected (in particular self.sat_sequence_len appears to differ then).
I guess a quick solution would be to change the default value of the min_sat_delay_minutes? Or remove the parameter all together as it is hard-coded in ocf_datapipes/training/pvnet.py; or make it configurable there.
Hi @tomasvanoyen, thanks for the issue. I agree this should be changed. I'll update the default and also add to datapipes so the delay can be configured
Describe the bug
Some default | hard-coded values are not consistent between
scripts/save_batches.py
(making batches) andrun.py
(setting model dimensions). In particular, runningscripts/save_batches.py
in non-production model sets thesat_delay
to 30 minutes (hard-coded, which is reasonable). However, running thereafterrun.py
provides as defaultmin_sat_delay_minutes: Optional[int] = 15
, leading (I believe) to a mismatch in dimensions expected (in particularself.sat_sequence_len
appears to differ then).I guess a quick solution would be to change the default value of the
min_sat_delay_minutes
? Or remove the parameter all together as it is hard-coded in ocf_datapipes/training/pvnet.py; or make it configurable there.Regards,
Tomas