🎉Community Raffle - Win $25

An exclusive raffle opportunity for active members like you! Complete your profile, answer questions and get your first accepted badge to enter the raffle.

Join and Win

Backup and retrieve modeling / Training status ?

Fred12

hi,

sometimes a training with grid optimization and different parameters for models (SVM, Tree, etc.) takes several days to weeks... but sometimes it happens that current is cut or something happens because of other problems... then I have to try out everything from beginning... now is it somehow possible to store the state of the machine while it is training the model with the parameters somehow intermediary? I mean like every 1 hour or so make a kind of "backup" of the training status, save it somewhere, and when it crashes, start from the backup and not start from square one?

is there any possibility to do that?

Find more posts tagged with

AI Studio

Accepted answers

bhupendra_patil

Hi Fred,

Please see the video below.

If something like this may be useful for you. I have also attached the sample process/workflow here

save intermediate results.zip

All comments

MartinLiebig

Hey Fred,

what abou this idea: You store a model every hour. If you need the parameters for another learner, you can use clone Parameters to transfer it.

Some branch operator with date_now() and %-function should be fine to do the "every hour" thing

~Martin

bhupendra_patil

Hi Fred,

Please see the video below.

If something like this may be useful for you. I have also attached the sample process/workflow here

save intermediate results.zip

MCs

This "Save Intermediate Results" approach and solution was very helpful and came just in time for me, too

Thanks @bhupendra_patil !

Chill

Fred12

thanks, that's pretty cool

Fred12

I have an additional question to that...

sometimes if I have too much parameters, its hard to save all the results and parameter models into a file, as its getting too much...

is it somehow possible to do a saving like say every 10 minutes or every hour with some sort of timer or so? then save the model and results of the performance..?