These are chat archives for beniz/deepdetect

22nd
Aug 2016
Michael Holcombe
@themolco
Aug 22 2016 00:56
this is probably a less than intelligent question: where does the docker version output the files? i can't seem to find any of the server files?
Matt Zeiger
@mattzeiger
Aug 22 2016 03:10
This is more of a docker question than one related to this project specifically, have you read up on how to share folders with your docker instance https://docs.docker.com/engine/tutorials/dockervolumes/
sunatthegilddotcom
@sunatthegilddotcom
Aug 22 2016 04:16
@themolco you can login to docker container with command docker exec -it docker_container_id bash
@themolco the default models are under /opt/models, other files are under /opt/deepdetect
Varun Jewalikar
@neo01124
Aug 22 2016 13:37
hi
Is it recommended to run multiple dede processes on different ports ?
on the same machine
Emmanuel Benazera
@beniz
Aug 22 2016 14:03
not really, unless you have something or some reason in mind ?
Varun Jewalikar
@neo01124
Aug 22 2016 14:21
ok, want to have some kind of production vs test setup on one machine
Emmanuel Benazera
@beniz
Aug 22 2016 14:22
yes, that's one of the settings where you'd want to have separate instances
Varun Jewalikar
@neo01124
Aug 22 2016 14:22
sometimes need to break training calls or there are crashes which would be on the test port and the production just sees the trained/finetuned models
Emmanuel Benazera
@beniz
Aug 22 2016 14:23
there shouldn't be crashes. We try to make it robust even to user mistakes, though Caffe can be difficult to recover from. But any time you can prove to crash the server deterministically, and not for lack of memory, you should report it as an issue
Varun Jewalikar
@neo01124
Aug 22 2016 15:15
Is there a simple way to get the previously running services back after a crtl-c or a crash ? Right now I just recreate them from the python interface or with curl calls one by one.
Emmanuel Benazera
@beniz
Aug 22 2016 15:16
no, you need to make a script and execute it. But there shouldn't be a crash :)
Varun Jewalikar
@neo01124
Aug 22 2016 15:42
ok got it