Default training in MR FIX is non-optimal
glasserm opened this issue · comments
HCPpipelines/ICAFIX/hcp_fix_multi_run
Line 745 in 3a6406c
We are currently defaulting to hp2000 SR+FIX training, when that is not optimal. We should default to the MR+FIX training file, as that is more generic (has MR and SR FIX data, HCP-YA scanner and Prisma, rest and task).
If we change hcp_fix_multi_run now, there is a risk of users getting different behavior for the same calling options, without them knowing it. Maybe we should add a warning instead?
The current behavior is a bug and is causing problems. I do think we should publicize the change, but we do need to make a change.
Where is the MR FIX training file?
Also, are you suggesting that for the batch file, when configured for SR FIX, that it should use the MR FIX training anyway, or continue to do what it does now?
The MR+FIX training file is included in the FIX distribution and is named:
fix1.06.15/training_files/HCP_Style_Single_Multirun_Dedrift.RData
It should work fine for SR+FIX too, so we could change it to be the default for everything or just MR+FIX. Since we recommend MR+FIX in most cases, this will cover most users anyway.
I would leave the default for SR-FIX as it is.
What would you have it do if the fix distribution is too old to have the MR training data?
Error and tell them to update.
This should be fixed now, though it needs to be tested.
Thanks Tim