use existing" classifier function

[Samuel Levine]

Hi Stepher,

Quick question regarding the "use existing" classifier function. 

If I create a source using an existing classifier *lets call the existing classifier "X"* and then train the new source with lets say 100 new images. Does this training lead to X being changed, or does it create a copy of X that is built off of. 

Curious on whether potentially adding poor training data in a new source could impact previous source performance. 

Thanks,

Sam

[Stephen Chan]

Hi Sam,

That does not lead to X being changed, nor does it build a new classifier off of X. You either select a specific classifier X with 'Use existing', or you use the 'Train' option to use a classifier trained on your 100 images instead. But with 'Use existing', it doesn't work together with your source's data.

That's how it goes for now at least. Maybe in the future we'll be able to do something more advanced.

[Jordan Pierce - NOAA Affiliate]

Also, if a user wants to create a new source and they select Classifier mode "Use existing", but then switches back to "Train", it locks them out:

[Stephen Chan]

Hi Jordan,

Ah yes, I think that happens if the 'Classifier global ID number' field is still filled in even though you have 'Train' selected. So the way to get un-stuck would be to switch back to 'Use existing', clear out that ID number field, and then switch back to Train. Though obviously the error message can be confusing in this case.