Syncing of AI Training

Assuming one has developed a log of ‘AI’ training for a given database on a given Mac … does the ‘See also and Classify’ improvements sync across databases ? Should one perform the training only on the ‘Master’ Mac … for instance… I have a ‘desktop’ Mac as well as a ‘Laptop Mac’. If I do not want to double the effort for the AI training, should I only perform it on one of the two computers ?

Where is the ‘Training’ data stored… how is it back-up ?

I can’t find any reference to “log of AI training” in the “DEVONthink Handbook.” Perhaps my bad. Please elaborate.

What is your concern? What problem are you fixing our seeking what opportunity? Do you notice one or more of your computers taking up too many CPU cycles or something?

I don’t know where the “AI” data is stored, but my hunch is, based on thinking about it, that it is stored in the database and therefore is probably synced but maybe not, but others can comments on that. Backups of any separate files related to AI not in the database (if that is the case) are wherever you put your full backups (TimeMachine or equiv).

I got autocorrected from ‘A Lot’ to ‘A Log’ … I meant ‘a lot of AI Training Data’

Development could also chime in here, but the metadata for a database is stored in the database package. AI is done outside the databases and is updated dynamically, so that data isn’t synced between machines.

@Bluefrog, please confirm that a machine backup, e.g. TimeMachine or similar, which includes the DEVONthink database will have the “AI” data will backed up to wherever the user sends backup files. @fmenard123 asked about that also. If not, what folders contain this for backup purposes?

I don’t believe there are files for the AI, though perhaps there is cached data.
I believe the classifications and See Also data is generated on demand, dynamically.

@cgrunenberg would have to shed further light on this.

Indeed, if the see also and classify is generated on demand dynamically, then there would be no hard training to store and therefore in possible need of sync. Looking forward to know!

The only thing that DEVONthink needs is the data and that’s synchronized. Everything else is handled automatically.

In the end, I just need the confirmation that training done on two computers on a synced database is complimentary rather than mutually exclusive. F.

There’s no explicit training, DEVONthink just uses the available data which is synchronized. Meaning that the same documents in the same groups on both computers should produce the same results.


What an interesting thread! Pretty cool!

Aha, that’s a great tidbit! So the AI is not “training” on the data. But what is it basing its suggestions then on? (Internal) Rules?

Just the contents of the database and how it’s organized.